YAMAHA

AUTHORIZED PRODUCT MANUAL
MULTITRACK CASSETTE RECORDER

OPERATING MANUAL

Congratulations on your choice of the New Yamaha MT1X Multitrack Cassette Recorder. The MT1X is a compact multitrack recorder with a recording mixer, and is equipped with numerous versatile functions. Using conventional cassette tapes, the MT1X makes it easy for you to produce high quality multitrack recordings. Besides use as a multitrack recorder, the MT1X can also be used as a PA mixer for small performances, as well as for editing soundtracks for videos. To take full advantage of the outstanding array of features, and for years of trouble-free operation, we urge you to thoroughly read this operating manual. After reading, keep it in a handy place for reference.

CONTENTS

BEFORE OPERATION. PLEASE NOTE THE FOLLOWING PRECAUTIONS. THE DIFFERENCE BETWEEN TRACKS AND CHANNELS. WHAT IS A MULTITRACK CASSETTE RECORDER?. THE CONTROLS AND THEIR FUNCTIONS.. MIXER SECTlON. RECORDER SECTION METER AND MONITOR SECTION. CONNECTOR SECTION. CONNECTION EXAMPLE. ABOUT CASSETTE TAPES. ATTACHING THE STRAP. WHEN USING THE BATTERY PACK. MULTITRACK RECORDING TECHNIQUES. ONE EXAMPLE OF A MULTITRACK RECORDING PROCESS. BEFORE RECORDING. MULTITRACK RECORDING. SYNC-RECORDING.. EDITING VIDEO SOUNDTRACKS. MAINTENANCE. BLOCK DIAGRAM. SPECIFICATIONS. INTRODUCTION TO THE ACCESSORIES. SERVICE. 38

BEFORE OPERATION

PLEASE NOTE THE FOLLOWING PRECAUTIONS:
ABOUT CASSETTE TAPE This unit is designed to be used only with Chromeposition tape, and will not work properly with Ferrichrome tape formulations. CrO tape (Bias: HIGH; EQ: 70s) should be used. Also, the use of C-120 tapes is not recommended because they exhibit poorer performance, and can be the cause of equipment failure. ABOUT dbx In order to get proper sound reproduction, set the dbx switch ON when playing back tapes recorded with dbx on, and keep it OFF when playing back tapes recorded without dbx. *dbx and the dbx mark are trademarks of dbx incorporated. *The dbx system has been manufactured under license of dbx Incorporated. USING THE AC ADAPTOR Please use the AC adaptor supplied with this unit. Other AC adaptors may vary in plug dimensions, polarity, or output voltage, and their use with this unit could cause damage. CAUTIONS FOR THE AC ADAPTOR Do not plug or unplug the AC adaptor with wet hands - you could receive a very dangerous shock. To avoid shorts or cord breakage, do not pull the plug out of the AC outlet by pulling on the cord. Be sure to grasp the plug itself and pull it out. When leaving home for an extended period of time, or when the unit will not be used for a long time, unplug the AC adaptor. NOTE: The AC adaptor has been designed for use with 120V or 220-240V AC, and must not be used in areas with different voltage. FCC CERTIFICATION (USA) This equipment generates and uses radio frequency energy and if not installed and used properly, that is, in strict accordance with the manufacturers instructions, may cause interference to radio and television reception. It has been type tested and found to comply with the limits for a Class B computing device in accordance with the specifications in Subpart J of Part 15 of FCC Rules, which are designed to provide reasonable protection against such interference in a residential installation. However, there is no guarantee that interference will not occur in a particular installation. If this equipment does cause interference to radio or television reception, which can be determined by turning the equipment off and on, the user is encouraged to try to correct the interference by one or more of the following measures: Reorient the receiving antenna. Relocate the computer with respect to the receiver. Move the computer away from the receiver. Plug the computer into a different outlet so that computer and receiver are on different branch circuits. If necessary, the user should consult the dealer or an experienced radio/television technician for additional suggestions. The user may find the following booklet prepared by the Federal Communications Commission helpful: How to identify. and Resolve Radio-TV interference problems. This booklet is available from the U.S. Government Printing Office, Washington, DC 20402, Stock No. 004-000-00345-4. PRECAUTION AGAINST LIGHTNING In the event of lightning or electrical storms, unplug the AC adaptor as soon as possible to avoid potential damage. DO NOT OPEN THE CABINET To avoid electrical shock or damage to the unit, do not open the cabinet and tamper with the parts or circuits inside. CONNECTING OTHER EQUIPMENT Make sure the power switch is OFF and the input fader is all the way down when connecting other equipment. M0VING THE UNIT To prevent shorts or breakage, make sure all connection cords have been removed from the unit before moving it. CLEANING THE CABINET Do not clean the unit with benzene or other powerful solvents, and avoid the use of aerosol insecticides near the unit. Clean the unit only with a soft, dry cloth.

THE DIFFERENCE BETWEEN TRACKS AND CHANNELS The words track and channel are often confused. In order to properly operate this unit, it is necessary to understand the meanings of these terms. TRACK: The band on the tape itself where a certain signal is recorded. The tape inside a cassette has four different tracks, enabling the recording of four distinct signals. For conventional recordings, there are two tracks (stereo left and right) on each side of the tape. CHANNEL: The route of a signal input or output. In the input side, this unit has four INPUT channels and two AUX channels. The output side consists of one stereo channel (made up of two mono channels) and an AUX channel.
WHAT IS A MULTITRACK CASSETTE RECORDER? The difference between the MT1X multitrack cassette recorder and a conventional stereo cassette deck is shown below.
CONVENTIONAL STEREO CASSETTE DECK
The diagram shows how a conventional stereo cassette deck records and plays back music. The four tracks on a cassette tape represent the left and right (for stereo) sound for each side of the tape. The MT1X uses these four tracks for single-direction recording and playback on only one side of a cassette tape. Conventional stereo cassette recorders always record and play back in the same mode, with the tape side (direction) determining which two tracks will be used. These recorders do not allow separate selection of tracks for recording and playback. Multitrack recorders, however, allow you to record or playback tracks separately as you choose. This enables a greater degree of recording and playback freedom not possible with conventional cassette recorders.
MT1X MULTITRACK CASSETTE RECORDER
THE CONTROLS AND THEIR FUNCTIONS
This section explains the names and functions of all the knobs, sliders, and switches for the mixer, recorder, meter/ monitor, and connector sections. Familiarize yourself with them in order to take full advantage of the MT1Xs versatile functions.

MIXER SECTION

INPUT SELECTOR SWITCHES These three-position switches are provided for each channel. Position them according to the operation to be performed. MIC/LINE: Set this switch to the proper position when the output of a microphone, keyboard instrument, or electric guitar is connected to the input jack on the front panel. Be sure to set the switch to this position OFF: when the channel is not being used, or when you dont want to playback material already recorded on the track. Although sliding the input fader to the O position will stop the signal, its a good idea to also set the switch to OFF.

Set the switch to this position to playback material which has already been recorded on this channel. Channels 14 correspond to tracks 14 on the tape.
GAIN CONTROLS The controls adjust the input level of the channel to match the output level of a microphone or instrument connected to input jack Control from -10dB to -50db is possible. Adjust the output level of the microphone or instrument as outlined in its instruction booklet.
SOUND CHARACTERISTICS OF THE EQUALIZER AND VARIOUS MUSICAL INSTRUMENTS
INPUT FADERS These controls adjust the volume of the signal input, and send it to the equalizer. Each control is used for setting the sound level of its channel when mixing it with the signals of other channels. Position 7 on the scale is considered ideal for the lowest noise and distortion characteristics. -----Normal frequency ----- Harmonic sound components If accurate and comprehensive sound equalization is required, connect a graphic equalizer or a parameteric equalizer between the sound source and the input jack. When recording material that you will intend to pingpong (see Ping-ponging on page 25), later, give the input somewhat of a high boost with the Hi control to help preserve the high frequency response when the track is re-recorded. PAN (PAN POT) CONTROLS After volume level and equalizing, the input signal is sent to these controls. During mixdown (see Mixdown on page 31), each control helps determine the acoustsic position of the signal in regards to the stereo field. Turning the control all the way to the left puts the signal all the way over to the left side of the stereo soundspace; turning the control to the right sends the signal towards the right. At dead center, the signal comes out equally from the left and right channels.
Be sure to set the control to O for channels not being used. EQUALIZER CONTROLS These controls are used to adjust the tonal characteristics of the input signal, or the channel output during playback of previously recorded material. The LO (low) controls adjust the frequencies centering around 100Hz, while the HI (high) controls adjust the frequencies centering around 10KHz, with a 10dB boost or cut range for both controls. Use of these equalizer controls will help you to get the type of sound you desire, and allow you to bring the sound forward, clean up unclear sounds, and push down sounds at annoying frequencies. In order to properly use these equalizers, its important to understand the frequency response characteristics of various musical instruments. This is particularly true when trying to change the sound of a certain instrument, because you should know that instruments harmonic sound components as well. For example, the normal frequency range of a bass drum is between 50Hz and 150Hz. To bring out this sound so you can feel it better, the LO (low) control (which centers on the 100Hz frequency band) can be moved up a little. But the harmonic sound components are around 10KHz, so the HI (high) control should also be moved up a little to achieve the proper sound profile of the bass drum.

These controls are also useful in ping-ponging (see Pingponging on page 25).
AUX CONTROLS The MT1X is equipped with an AUX SEND jack and two (left and right) AUX RETURN jacks When special acoustic effects are desired on a certain channel, reverbs or delay effects can be connected between these jacks to provide only the desired effect to the desired channel. Amplified monitor speakers can also be connected to the AUX SEND jack. Each AUX control adjusts the sources connected to the AUX SEND jack in the following manner.
CONNECTING MONITOR SPEAKERS
Performers or sound mixers can control the level balance of the four channels (instruments) with the AUX controls level adjusted by the AUX MASTER SEND control

, with the total output

MASTER FADER This controls the level of all the input faders, as well as the final level of the effected signal of the AUX RTN control (the recording level and the sound mixed through the stereo mix buss. The output level of the ST OUT jack at mixdown) and the recording level during ping-ponging are also adjusted with this control.
Set the control input faders at 7 for best results.
AUX MASTER SEND CONTROL This control adjusts the level of the effect-mixed signals from each channel (adjusted by each AUX control as well as the AUX signal for monitoring use. The final output is through the AUX SEND jack AUX RETURN CONTROL This control adjusts the input level of effects or submixers connected to the AUX RTN jack. The level of effect in relation to the sound can be set with this control.
SYNC SWITCH Normally left in the OFF position, this switch should be turned ON if this unit is to be used for synchronized operation with MIDI products like synthesizers and rhythm machines. Synchro operation is explained in the or in the Sync-Recording secsection on Sync jack tion on page 33. Power Indicator This indicator lights when the power switch rear panel is turned on. on the
RECORD SELECT SWITCHES These switches are used to choose the signal to be recorded. The upper left switch is for track 1, the upper right switch is for track 2, the lower left switch is for track 3, and the lower right switch is for track 4. When the track is not to be recording, set the corresponding switch to the OFF position. Switch ON only those switches corresponding to the tracks which are to record. The panel indications for L and R correspond to the stereo left and right signals, whereas 1, 2,3and 4 correspond to the signal from the 1, 2, 3, and 4 input channels. Those signals are recorded on their respective track when the switches are in position. NOTE: Tracks 1 and 3 cannot be recorded on the right stereo signal, and tracks 2 and 4 cannot be recorded on the left stereo signal. REC INDICATOR Recording status is indicated in the following three ways: No indication: All tracks 14 are not recording. Flashing: All tracks 14 are in recording standby mode. By pushing only the REC switch the tape is put into the recording standby mode. Indication ON: All tracks 14 are recording, or in the REC pause mode. To resume recording during REC pause, press the PAUSE switch REC SWITCH When this switch is pressed, the PLAY switch also moves, and the unit goes into the recording mode. However, if the RECORD SELECT switches for all tracks 14 are switched OFF, nothing will be recorded. NOTE: When the REC switch is pressed down, noise occurs which is recorded on the tape. In order to prevent this, we recommended the use of the Press the PAUSE switch first, PAUSE switch then press the REC switch. To start recording, press the PAUSE switch again to shift out of the REC pause mode and into the recording mode. PLAY SWITCH Press this switch for playback. However, if the input selector switch (1) of a track is not in the TAPE position, the sound will not be heard on the stereo buss. REW SWlTCH (REWIND) Use this switch to rewind the tape. Pressing it when the MT1X is in the PLAY mode enables you to hear the sound of the tape while it rewinds. This feature is useful for finding the beginning of a song or other recorded material.

FF SWITCH (FAST FORWARD) Use this switch to quickly advance the tape forward. Pressing it when the MT1X is in the PLAY mode enables you to hear the sound of the tape while it is moved forward. This feature is useful for cueing up the start of a subsequent song or other recorded material on the tape. STOP SWITCH Press this to stop tape run. PAUSE SWITCH Press this switch to momentarily stop playback or recording in progress. Press it again to restart. dbx SWITCH Ordinary cassette tapes dont have sufficient dynamic range (the level difference between the softest sounds and the loudest peaks) to adequately record highly dynamic music. If the dbx switch is put ON during recording, highly dynamic music signals can be adequately handled, while the hiss noise inherent to cassette tapes is kept down below the range of human hearing. If the dbx switch is kept ON during recording, it must also be kept ON during playback. During recording or playback, this control can be used to vary the tape running speed from +10% to -10%. The pitch of voices or musical instruments also varies in proportion to tape speed. Under normal conditions, the control should be in the center position. When overdubbing (playing back a recorded passage while recording new material on a different track) the pitch of the previously recorded material can be altered to match the new material if necessary. This feature can also be used to obtain certain special effects during recording. TAPE COUNTER This displays the amount of tape run. RESET SWITCH Press this switch to reset the tape counter to 000. Pressing this switch at the start of recording, or at the beginning of a song, makes it easy to cue up the selection from the start. ZERO STOP SWITCH If this switch is set ON during rewinding, the tape will stop when the tape counter reaches 999. During multitrack recording, this feature is convenient for repeated playback or recording operations after rewind.

PITCH CONTROL

METER AND MONITOR SECTION

MONITOR LEVEL CONTROLS When setting the PHONES SELECT Switch to the MONITOR position, these level controls are used for each track to achieve a level balance for easy monitoring. Use these controls freely and independently to maintain a desired level balance during overdubbing operations, when the addition of a new signal changes the volume.
MONITOR PAN CONTROLS When setting the PHONES SELECT Switch to the MONITOR position, use these pan controls for each track to achieve the desired stereo positioning for each track. Use these controls freely and independently to maintain the desired stereo position balance during overdubbing operations, when the addition of a new signal changes the stereo image. PHONES CONTROL This control adjusts the volume of the headphones (See page 9).
CONNECTOR SECTION FRONT PANEL
INPUT JACKS These four jacks are for the connection of microphones or electric and electronic instruments. With a high input impedance of 10K ohms, and a specified input level range of -10dB to -50dB, a wide variety of instruments and microphones can be used. When directly connecting an electric guitar, the use of an special electric guitar preamp to increase the input level will assure recordings of better sound quality. REAR PANEL
PHONES JACK Plug a set of headphones into this jack for monitoring. Please use headphones rated from 8-0 ohms for best results. PUNCH IN/OUT JACK By connecting the optional FS-1 footswitch to this jack, you can control punch-in/punch-out operations. by foot.

POWER SWITCH

This switch turns on the MT1X. When switching the unit on or off, make sure that the Input Faders and the AUX RTN Control are at the 0 position.

DC IN JACK

Connect the supplied AC adaptor to this terminal. To prevent damage, use only the AC adaptor supplied with this unit.

SYNC IN/SYNC OUT JACKS

These jacks are used during synchronized operation with MIDI-equipped instruments. Use the optional YMC10 MIDI Converter to connect the instruments through these jacks, and set the SYNC switch on the mixer section to ON. For a detailed explanation, refer to SyncRecording on page 33.

TAPE OUT JACKS

These jacks directly output the signal of each track. During playback, the signals of the tracks being played are output. During recording, the signals of the tracks being recorded are output. Since the output levels cannot be adjusted, set the volume by adjusting the output levels on the instruments themselves. These jacks can be conveniently used in the following ways: Another 4-track recorder can be connected for direct dubbing of all four channels. An external mixer can be connected for mixdown.

ST OUT JACKS

The mixed signals of each channel (and each track) are output in stereo signal through these jacks. Since these jacks output the final mix, a stereo cassette deck can be connected. These jacks can also be used as follows. The MT1X can be used as a sub-mixer, with the output sent to a main mixer through this jack. A stereo amplifier or powered monitor speakers can be connected through this jack.

ATTACHING THE STRAP

The MT1X can be easily carried with the supplied carrying strap. Heres how to attach it. Push on the slit to open the stopper, and hang it on the peg. Slide the strap to the desired position and lock the stopper in place.
WHEN USING THE BATTERY PACK
With the optional PA11 Battery Pack, the MT1X can be operated by batteries in places where there is no AC outlet available. Heres how to set it up. PUTTING IN THE BATTERIES Slide off the battery cover on the bottom of the battery pack. ATTACHING THE BATTERY PACK Align the battery pack on the left side of the MT1X.
Slide it on, and tighten the battery pack mounting screw with a coin or screwdriver. The battery pack is now firmly attached. Insert 10 C batteries as shown. Put 7 on the bottom Then put 3 on top NOTES:
Replace the battery cover.
When using the battery pack as a power supply, remove the AC adaptor. The battery pack can only supply power to the MT1X when it is properly attached. The MT1X was designed for indoor use. Avoid using it areas of high heat or humidity, or in dusty places. If the battery pack isnt going to be used for an extended period of time, remove the batteries inside. Battery life: about 2 hours during 2 channel recording with a headphone output of 10mW + 10mW.
MULTITRACK RECORDING TECHNIQUES
Before you try to attempt a multitrack recording on your own, its absolutely essential that you understand the function of all the controls, switches, and connectors in each section. In addition, you should spend an adequate amount of time to familiarize yourself with the block diagram on page 35. It may appear hard to understand at first, but after carefully looking it over, youll find that its not only easy to fo!low, but quite useful in understanding the various signal flows involved in using the CMX1. The numbers on the block diagram for the controls, switches, and connectors correspond to those used in the section titled The Controls, and Their Functions. ONE EXAMPLE OF A MULTITRACK RECORDING PROCESS Multitrack recording is usually used to record a rhythm section, with overdubbing and ping-ponging operations assisting in mixing the parts of the various musicians in the proper balance. Finally, the tape is mixed down to produce a stereo master tape. These are the steps in our example: Record the drums on track 1 Record the bass on track 2 Record the rhythm guitar on track 3 Ping-ponging tracks onto track 4 (freeing tracks 13) Record the keyboards on track 1 Record the lead guitar on track 2 Record the vocals on track 3 Mixdown tracks to produce a stereo master tape mere instant, its not a problem. However, if theyre peaking out for more than a second or two, then distortion may become a problem. Its also important to remember that distortion at lower frequencies is less apparent than distortion at higher frequencies. dbx SYSTEM Keep the dbx switch ON to expand dynamic range and to reduce inherent tape noise. STEREO POSITIONING Its important to think about the acoustic position of all the instruments well before you start your multitrack recording. You should take into account a certain amount of noise caused by ping-ponging and mixdowns planned for later on. Heres one example of acoustic positioning. Set the bass drum and the snare drum center, with the tomtoms and high hat set off to either side to bring out the stereo effect. The bass and other heavy instruments should be in the center, with the keyboards to the left and the guitar to the right. Solo instruments and voices should span both right and left. Solo instruments with a stereo output can have their left channel connected to a delay machine, while the right channel is recorded directly. You can probably think of many other different ways to arrange the soundstage. EQUALIZATION AND EFFECT PROCESSING Equalization and effect processing are usually added at the ping-pong and mixdown stages. In multitrack recording, these types of signal processing can be decided on later, and employed to any degree necessary. However, the MT1X is limited in the number of effects which can be used during mixdown, so its best to use them during the initial recording stages. Monitoring BEFORE RECORDING RECORDING LEVEL In making a good recording, the most important step is setting the ideal recording level. If the level is too low, the recording will contain a lot of noise and hiss; if the level is too high, the recording will sound distorted and unclear. Set the recording level at a fairly high level, but not so high as to result in any noticeable distortion. The CMX1 is equipped with peak level meters which show the level of each track, as well as the level of the stereo output signal. Use these meters to help you set the ideal recording level, because the human ear has difficulty in detecting distortion immediately. If the level meters peak out (show the maximum reading) in a In addition to circuits for signal recording, this unit also features a separate monitor circuit to allow the performer to monitor the levels and positioning of the recording in progress through a pair of headphones. In this case, set the PHONES SELECT switch to the MONITOR position. Adjust the volume level and stereo positioning of each track with its MONITOR LEVEL and MONITOR PAN controls. In addition, powered monitor speakers can be directly connected to the ST OUT jacks or the AUX SEND jack, though this makes it impossible to use these jacks for external recording or effects. Using speakers during recording off lines presents no problems, but when microphones are used, feedback can result when the microphone picks up sound from the speakers. In this case, monitoring through headphones becomes absolutely necessary.

Plug a pair of monitor headphones (rated 8-40 ohms) into the PHONES jack. When using an effect, connect it between the AUX SEND jack (input) and either of the AUX RTN jacks (output).
4. Setting the monitor and meter sections Set the PHONES SELECT switch to the STEREO position. Set the PHONES volume to around 7. Make sure the METER SELECT switch is in the 4 TRK position. 5. Adjusting the input level Set all of the input switches to the MIC/LINE position. Set the MASTER fader to 7. Set the PAN controls for all channels between the center and the extreme L" position, as shown.
Push the input fader for channel 1 up to 7. When the drums start playing, slide the gain control for channel 1 towards the MIC, direction, stopping when the +3 indicator on the far left level meter flashes occasionally on the sound peaks. Pull the input fader for channel 1 back down to 0. -The proper input level for channel 1 is now set. Set the input fader for channel 2 to 7. While the drums are being played, slide the gain control for channel 2 towards the MIC direction, stopping when the +3 indicator on the far left level meter flashes occasionally on the sound peaks. Pull the input fader for channel 2 back down to 0: -The proper input level for channel 2 is now set.Set the input levels for channels 3 and 4 the same way.
*Explanation diagram for steps 6. Adjusting level balance and equalization characteristics Adjust channel faders 1 ~4 to achieve the desired recording level balance. Adjust the equalizers for 1~4 to obtain the desired sound character for each individual channel. (If youre thinking of ping-ponging these tracks afterwards, its a good idea to add a little boost on the HIGH EQ because high frequencies can be slightly diminished during the ping-pong re-recording process). Set the effect level for each channel with the AUX controls. Then, adjust the overall mix of effect signal to input signal with the AUX MASTER SEND control. If necessary, go over steps several times to until everything is just right. Adjust the master recording level with the MASTER fader, setting it at the point where the +3 indicator on the far left level meter flashes occasionally on the sound.peaks.
7. Recording Push the PAUSE switch to start recording. Just before the musician starts playing, be sure to count out loud to help you get the timing right on the other tracks during overdubbing later on. When the music sequence is over, press the STOP switch to stop the recording. Then, press the REW switch, and the tape will rewind to a point just 999 on the tape counter before the beginning and stop. Drum Recording Completed
Track 1 Track 2 Track 3 Track 4 20
Make sure this switch is in the 4 TRK position.
Set this so you can monitor track 1.
Press to check the recording on track 1, then rewind. to

*Explanation diagram of steps
8. Post recording check Return all switches and controls to their normal positions. Set the PHONES SELECT switch to the MONITOR position, turn MONITOR LEVEL control 1 to 7: then turn the PHONES volume control to about 7. Make sure the METER SELECT switch is set to 4 TRK". Press the PLAY switch, and check the sound recorded on track 1 by headphones, and by looking at the level meter.
At this point, if the track is recorded properly and there doesnt seem to be any problems, press the REW switch and reset all the knobs and controls to their normal positions. Now its time to record the bass. If the recording is not to your satisfaction, you can rerecord the whole track. Or you can use the punchin/punch-out technique to record over a certain spot on the tape. Its explained on page 27.
RECORDING THE BASS GUITAR BY OVERDUBBING
Overdubbing is the playing back previously recorded tracks while recording a new instrument on a different track. With this technique, its possible for one musician to play many different instrumental parts on a single recording. If youre multitalented, multitrack overdubbing can clone you into your own group. Now, were going to record a bass guitar on track 2 to synch with, or match, with the drum part already recorded on track 1. There are two ways to record the bass: place a microphone in front of the bass amp, or run a direct line from the bass into the recorder. If youre after a really clear recording, direct line recording is the way to go. If youre using an effect of some sort, youll want to run a noise gate thru the final stage of the effect. This is true with all electrified musical instruments. Another idea is to use the Yamaha GC2020 comp/limiter. In addition to the compresser and limiter functions, the noise gate function can prove to be very convenient.
Signal Path when Recording the Bass Guitar
Bass guitar recording procedure
1. Connections Connect everything through input jack 2 as follows. If the GC2020 is being used, connect it between the amplifier and input jack 2. When not using a bass amp, the use of a direct box is recommended.
2. Setting the recorder Make sure the tape has been rewound to the 999 point on the counter. (This also goes for the rest of the steps.) Keep the ZERO STOP switch ON until mixdown. (This also goes for the rest of the steps.) Keep the dbx switch ON until mixdown. (This also goes for the rest of the steps) Set the RECORD SELECT position to 2, the REC indicator will flash to show that the bass guitar connected to input jack 2 will be recorded on track 2. Press the pause switch to start the recording. The REC indicator will light up completely. 3. Setting the monitor and meter sections Set the PHONES SELECT switch to the MONITOR position. Turn MONITOR LEVEL controls 1 and 2 to about 7: Set the PHONES volume level to about 7. Make sure the METER SELECT switch is in the 4 TRK position.

Connect the monitor headphones. Until the mixdown process, only use headphones and avoid using monitor speakers. (This also goes for the rest of the steps.) 23

Equalizer controls

Press the PAUSE button and adjust the monitor levels
Set to the MIC/LINE position Set by the reading on the level meter Push up to about 7 Set after setting equalization *Explanation diagram for steps to
4. Adjusting the Input level Set the input selector switch to the MIC/LINE position. Push input fader 2 up to about 7. Start playing the bass guitar, and slide gain control 2 towards the MIC direction, stopping when the +3 indicator on the level meter second from the left flashes occasionally on the sound peaks. 5. Adjusting the recording level and sound characteristics Operate the equalizer controls for channel 2 to get the desired tone. (If you plan to ping-pong this track later, boost up the treble a little bit with the HIGH EQ control.) Use input fader 2 to adjust the recording level according to the reading on the meter second from the left. 6. Adjusting the monitor sound Press the PAUSE switch to start the tape, and set the sound balance of the bass guitar and drums. If necessary, control the combined volume level with the PHONES volume control. Now, using MONITOR PAN controls 1 and 2, decide the left/ right stereo positioning of the two tracks. (During this, the bass guitar will be recorded on track 2.)
After you have adjusted the monitor levels and pans to your satisfaction, rewind the tape and set the recorder into the REC PAUSE mode. 7. Recording Press the PAUSE switch to start recording. While monitoring through headphones, the bass player should play along in synch with the drum track. When the musical segment is over, stop and rewind the tape. Bass Guitar Recording completed Track 1 Track 2 Track 3 Track 4
8. Post recording check Just press the play switch to check to see that the track was recorded properly. Set all switches and controls back to their normal positions.
RECORDING THE RHYTHM GUITAR Record the rhythm guitar on track 3 to synch with the drums on track 1 and the bass guitar on track 2. Recording preparations and operations are the same as when recording the bass guitar. If effects are being used, connect them just before the input jack. PING-PONG < PING-PONG RECORDING > After the rhythm section has been recorded on tracks 1 3, only track 4 remains as an empty, spare track. Since there are three more parts to be recorded, more tracks will be needed. The ping-pong technique (sometimes called bouncing, or track transfer, or collapsing tracks) shown here allows you to re-record these three tracks onto one track, thus freeing up tracks for other recording operations. You can also add other new parts during the ping-pong process. As long as there are empty tracks, you can ping-pong from one or more tracks to another as many times as you like. However, each time a track is ping-ponged onto another track, some degradation in sound quality occurs. Most noticeable is a loss of high frequency sounds, or treble. Therefore, its best to plan for only 1 or 2 ping-pong operations to preserve the sound quality of the instruments you record. Now, lets get started. Signal Path during Ping-pong Recording

SYNC-RECORDING

For synchronized operation with MIDI instruments, the optional YMC10 MIDI converter enables synchrooperation of the CMX1 and MIDI instruments such as the RX11, RX15, and RX21 digital rhythm programmers and the QX1 and QX7 digital sequence recorders. In this section, we will explain synchro-recording using synchro-operation techniques. Merits of synchro-recording Synchro-recording enables the use of digital sources such as rhythm programmers and sequencers during the first mixdown stages. Since these sources are recorded directly onto the master, it extracts the full sound quality, dynamic range, and superb S/N performance of these digital instruments. Operating the tape sync With the SYNC switch ON, press the PLAY switch and the FSK signal recorded on track 1 is sent to the YMC10, which converts it to the MIDI synchronizing signal and outputs it to the RX15. In this way, track 1 of the MT1X operates the RX15. FOR THIS OPERATION, DONT FORGET TO SET THE RX15 SYNC SWITCH TO MIDI. Now, tracks 2-4 can be used for overdubbing. Connect as shown below for mixdown.
In order to work the tape sync, the MIDI synchronizing signal must be converted to an FSK (frequency shift keying) signal first. This is because MIDI transmits information at an extremely high maximum speed of 31.25 K baud per second. Therefore, the use of analog instruments is impossible. By using the MIDI converter, the MIDI synchronizing signal is converted to an FSK signal that analog instruments can handle. An example of synchro-recording using the RX 15 rhythm programmer Set the RX15 to create the desired rhythm effects. Connect the RX15 in the following manner. Start the tape, mixdown the sound from tracks 24 and the RX15s drum sounds input through the AUX Left and Right jacks, and record it with a stereo tape deck.
After putting the MT1X into the REC PAUSE mode, turn the SYNC switch ON. In this condition, both the tape will start and the RX15 will start playing when the PAUSE switch is pressed. Heres how it works. The YMC 10 converts the MIDI synchronizing signal from the RX15 to FSK signal, which is recorded on track 1 of the MT1X. In order to operate the RX15 by the FSK signal recorded on track 1, connect everything like this:

EDITING VIDEO SOUNDTRACKS
Most people will agree that the sound recorded by the video cameras microphone just isnt enough for a good music video. Using the MT1X to edit down a good soundtrack for your video is a great idea, and will result in a video that sounds surprisingly professional. Youll find this capability useful to produce a promotional video for your group. In the following example, well show you how to make a soundtrack that includes the sound recorded by the video cameras microphone, narration, background music, and sound effects. Editing Procedure Playback the video several times in order to create a good, tight scenario. If youre going to edit the video footage, do it first. Use track 1 to record the sound recorded by the video cameras microphone. Track 2 is for recording the narration. While watching the video and monitoring track 1 with headphones, record the narration with a microphone. Overdub the background music on track 3. If this music is in stereo, use tracks 3 and 4. If its just mono, track 3 will suffice. Sound effects can be recorded on track 4. Mix down the sound from tracks 1-4 and record them on the video soundtrack using the video decks overdubbing function. NOTE: This example is when editing the soundtrack of a monaural video deck with an overdubbing function.

MAINTENANCE

As a good general rule, the tape heads, pinchroller and capstan should be cleaned before each recording, thus ensuring the best audio quality. After the deck has been used for a period of time, the heads, pinchroller, and capstan will become dirty. This increases noise and uneven rotation, leading to a deterioration in sound quality. Therefore, periodic cleaning and demagnetization is a must to preserve optimal audio performance. Use cotton swabs and alcohol or head cleaning fluid (available in most all audio stores) to clean the heads, capstan, and pinchroller. Keeping the heads clean is essential for good recordings. For demagnetization, use a quality head demagnetizer, and follow the instructions carefully.
Its important to keep all parts clean!

BLOCK DIAGRAM

NOTE: When the REC button is engaged, the panel switches can be used to individually order recording on only those channels with RECORD SELECT not switched OFF.

(201) 644-2332 or Eedie & Eddie on the Wire An Experiment in Music Generation
Peter S. Langston Bell Communications Research Morristown, New Jersey
ABSTRACT At Bell Communications Research a set of programs running on loosely coupled Unix systems equipped with unusual peripherals forms a setting in which ideas about music may be aired. This paper describes the hardware and software components of a short automated music concert that is available through the public switched telephone network. Three methods of algorithmic music generation are described.
Introduction Ten years ago, in order to experiment with computer-generated music, a researcher would have required a large computer and a great deal of special purpose equipment or would have had to settle for orchestrating the serendipitous squeaks and squawks of other equipment.0 In the last few years, advances in signal processing techniques and large-scale integration, combined with the proliferation of consumer music products have brought the cost of computer-controlled music hardware down to that of conventional musical instruments. Sound processing hardware is now reaching a level of availability comparable to that reached by text processing hardware ten to fteen years ago. In the next ten years it would not be unreasonable to expect intense activity in the area of sound manipulation software, with a revolutionary impact on industries that depend on sound. Sound can be used in many ways; the single most important use is as a communications medium. Spoken language is the most obvious of the techniques for communication via sound, and it is has been the subject of intense research for many years. Another use of sound is to create an experience that communicates on a non-linguistic level. Music incorporates both communication and experience but the line separating them becomes somewhat indistinct. What is it we enjoy in a piece of music? There are convincing arguments to the effect that music has both a vocabulary and a grammar, with different types of music having different vocabularies and grammars. Thus, the enjoyment of music becomes a learned skill, much like reading French or Latin, and new types of music appear to be gibberish until they are learned, e.g. Gamelan music is Greek to me. There are many interesting questions to be answered: What is the language of music? Does it have a grammar? What differentiates a jumble of notes from a piece of music? What are the semantics of music? Can you say The countryside is very peaceful or even I have lost my American Express card in
0 In 1965 I even wrote a compiler to turn music notated on punched cards into a program that, when run, would produce controlled radio interference and thereby play music on a nearby transistor radio.
music? Before we can hope to answer any of these questions we need to have a lot more data. The present project seeks to provide some data for questions about differentiating music from a jumble of notes through subjective evaluation of the output of programs that assemble notes by relatively simple rules. Hopefully we will be able to draw some conclusions from these data. For instance: a) If the program outputs are deemed musical then the rules used in the program are sufcient, b) If the outputs of programs with nonoverlapping sets of rules are deemed musical, then neither set of rules is necessary. The approaches used here for music generation deal only with syntactic (i.e. form) considerations; the programs have no methods for handling semantics. The semantics of music are assumed to involve an immense wealth of cultural and historical information (a.k.a. knowledge) that does not yet exist in any machine readable form. Understanding the semantics of music is no simpler than understanding the semantics of natural languages; for that matter, it would be easy to argue that music is a natural language, albeit often a non-verbal one. Concurrent with the project in music generation at Bell Communications Research is a project exploring the benets of interconnecting computers and telephone equipment [REDMAN85] [REDMAN86]. As a demonstration of both projects, an E&M trunk (a direct audio connection) on our experimental telephone switch was allocated to provide telephone access to the music hardware. Some programs were then written to present, in an entertaining form, examples of the music generated. (201) 644-2332 A block of 100 telephone numbers (644-2300 through 644-2399) have been allocated to an experimental telephone switch in our lab which has been dubbed BerBell (the switch itself is manufactured by Redcom, Inc., which has no direct relationship to B. E. Redman, logname ber, the principal investigator on the BerBell project). Aside from providing enhanced telephone service for approximately 40 people lines, (32 in-house extensions, two remote lines to residences, and several numbers to which participants can forward their home phones to use BerBells call management features), a number of experimental services are provided. Among these services are: 644-2300 644-2311 644-2312 644-2312 644-2331 644-2332 644-2335 644-2337 and others. Figure I shows the various hardware modules that form the system and their interconnections. Lines terminated with square boxes are EIA RS-232 serial connections (bidirectional). Lines terminated with simple arrowheads are audio connections (directional). Lines terminated with pentagonal arrowheads are Musical 1 Instrument Digital Interface (MIDI ) connections (directional). The remaining connections are either Ethernet lines (terminated with rectangles) or telephone lines (with no special termination symbol). The E&M trunk mentioned earlier connects the equalizer in the upper left corner of the gure with the Redcom telephone switch. When a call comes in to 644-2332 a process called demo2332 is spawned on the VAX 11/750 (yquem in gure I). This process rst connects the Dectalk speech synthesizer (Eddie in gure I) to the callers incoming line and then proceeds to introduce the demo, ask the caller to enter his/her telephone number

In addition, several pieces of sound processing equipment are in use:
The nal pieces of special hardware are those used to connect the Sun workstation to the MIDI instruments:
Using these peculiar peripherals and approximately 50 special purpose programs, it is possible to generate, record, play, and edit musical pieces with an ease that astounds conventional musicians.2 The appendix contains a list of the principal programs used. The total cost of this particular selection of hardware, not including the VAX, the SUNs, or the REDCOM telephone switch, was about $15,000 (based on list prices). In the last year, many prices have dropped and some items have been replaced by equivalent, less expensive equipment. The telephone demo, as described in this paper, could now be implemented on equipment costing less than $4,000. Music Generated by Optimized Randomness Riffology Schemes for harnessing random events to compose music predate computers by many years. Devised in the eighteenth century, Wolfgang Amadeus Mozarts Musical Dice Game gave an algorithm for writing music by rolling dice. Since that time, researchers at the University of Illinois, Harvard University, Bell Telephone Laboratories, and numerous other places have replaced dice with computers and turned disciplines such as signal processing, combinatorics, and probability theory toward the task of composing music [HILLER70]. The idea for the rst music generation technique used in the telephone demo did not come from these illustrious predecessors, however. It came from experiences as lead guitarist in several bands that performed improvisational music. One of the popular criticisms that could be levelled at another guitarist was that he or she just strings a lot of riffs together and plays them real fast. Of course, you were supposed to be 3 pouring out your soul and a little bit of divinely-inspired ecstasy instead. The principal objection to playing an endless succession of riffs is that it doesnt involve a great deal of thought, just good technique. That is to say, the syntax is more important than the semantic content. For this reason, an algorithmic implementation of riffology need not be hampered by its inability to manipulate semantics. The theme music for the video game ballblazer (tm, Lucaslm Ltd.), called Song of the Grid (tm), is generated by just such an approach [LEVINE84] [LANGSTON85]. The program runs in little memory on a small 8-bit processor (a 6502) connected to a sound chip that can make 4 independent sounds at once using square waves and pink noise. The riffology algorithm makes dynamically weighted random choices for many parameters such as which riff from a repertoire of 32 eight-note melody fragments to play next, how fast to play it, how loud to play it, when to omit or elide notes, when to insert a rhythmic break, and other such choices. These choices are predicated on a model of a facile but unimaginative (and slightly lazy) guitarist. A few examples should illustrate the idea. To choose the next riff to play, the program selects a few possibilities randomly (the ones that come to mind in the model). From these it selects the riff that is easiest to play, i.e. the riff whose starting note is closest to one scale step away from the previous riffs ending note. To decide whether to skip a note in a riff (by replacing it with a rest or lengthening the previous notes duration) a dynamic probability is generated. That probability starts at a low value, rises to a peak near the middle of the solo, and drops back to a low value at the end. The effect is that solos

These are tools for musicians, however, not replacements for them. I was never accused of this form of riffology myself; I always stuck to straight soul-pouring.
Figure II - Riffology from Song of the Grid start with a blur of notes, get a little lazy toward the middle and then pick up energy again for the ending. The solo is accompanied by a bass line, rhythm pattern, and chords which vary less randomly but with similar choices. The result is an innite, non-repeating improvisation over a non-repeating, but soon familiar, accompaniment. A version of Song of the Grid forms part of Eedies telephone demo. This nite version has a more standard structure; the accompaniment has a xed AABA pattern and a precomposed head (melody) is played the rst time through the pattern. The following improvisation uses one, two, and nally three interdependent voices. The repertoire of riffs has been increased by about 40% and now contains some famous phrases from early jazz guitarists who, being dead, could not be asked for permission (the original selection contained many riffs contributed by friends and used with their consent). Figure II is the beginning of one of the improvisations produced. The music generated by this algorithm passes the is it music? test; Song of the Grid makes perfectly acceptable background music, (better than that heard in most elevators or supermarkets) and even won lavish praise in reviews of ballblazer in national publications. However it doesnt pass the is it interesting music? test after the rst ten minutes of close listening, because the rhythmic structure and the large scale melodic structure are boring. It appears that for music to remain interesting it must have appropriate structure on many levels. Music Generated by Formal Grammars L-Systems In the nineteen-sixties, Aristid Lindenmayer proposed using parallel graph grammars to model growth in biological systems [LINDENMAYER68]; these grammars are often called Lindenmayer-systems or simply L-systems. Alvy Ray Smith gives a description of L-systems and their application to computer imagery in his Siggraph paper on graftals [SMITH84]. Figure III shows the components of a simple bracketed 0L-system grammar4 and the rst seven strings it generates. This example is one of the simplest grammars that can include two kinds of branching (e.g. to the left for ( and to the right for [). The name FIB was chosen for this grammar because the number of symbols (as and bs) in each generation grows as the bonacci series.

A 0L-system is a context insensitive grammar; a 1L-system includes the nearest neighbor in the context; a 2Lsystem includes 2 neighbors; etc. A bracketed L-system is one in which bracketing characters (typically [ and ], or ( and ), or others) are added to the grammar as placeholders that indicate branching, but are not subject to replacement.

Alphabet: Axiom: Rules:

{a,b} a ab b (a)[b] String a b (a)[b] (b)[(a)[b]] ((a)[b])[(b)[(a)[b]]] ((b)[(a)[b]])[((a)[b])[(b)[(a)[b]]]] (((a)[b])[(b)[(a)[b]]])[((b)[(a)[b]])[((a)[b])[(b)[(a)[b]]]]] Figure III - FIB, a simple bracketed 0L-system

Generation 6

L-systems exhibit several unusual characteristics and three are of particular interest here. The rst is that they can be used to generate graphic entities that really look like plants. Although the grammars themselves are quite mechanistic and have none of the apparent randomness of natural phenomena, the graphic interpretations of the strings generated are quite believably natural. The second characteristic is structural. The words of the grammar (the strings produced by repeated application of the replacement rules) have a ne structure dened by the most recent replacements and a gross structure dened by the early replacements. Since all the replacements are based on the same set of rules, the ne and gross structures are related and produce a loose form of self-similarity reminiscent of the so-called fractals (although none of the entities produced by graph grammars are fractal in the strict sense of the word). The third characteristic of L-systems is a property called database amplication, the ability to generate objects of impressive complexity from simple sets of rules (i.e. generating a lot of output from little input). We would like any composition algorithm to be signicantly less complicated than the music it generates. The generation of the strings (or words) in 0L-systems is strictly dened; every possible replacement must be performed each generation, or, to put it another way, every symbol that can be replaced will be. This means that the process is deterministic as long as no two replacement rules have the same left side. Note that in cases where the rewriting rules specify replacing a single symbol with many replaceable symbols (the typical case), growth will be exponential. Although generation of the strings is well dened, an additional interpretation step is required to express

FIB - gen 14 ab b (a)[b]

SMITH - gen 5 a b[a]b(a)a b bb
DRAGON - gen 5 a b[a[ba]] b b((b)a)c c cd
TERNARY - gen 6 a d[dbe](dce)e b d[daf](dcf)f c d[dbg](dag)g
SCRUB - gen 5 a c(ba(b))c[ba[b]] b c(be)c[bf] c cgg
Figure IV - graphic interpretations of some graph grammars these strings as graphic or auditory entities. Figure IV shows a very simple graphic interpretation of several grammars, including that of FIB.5 In this interpretation the symbol ( begins a branch at an angle of 22.5
Missing from the grammar descriptions are their alphabets ({a,b}, {a,b}, {a,b,c}, {a,b,c,d,e,f,g}, and {a,b,c,e,f,g} respectively) and their axioms (a for all of them).
to the left (i.e. counterclockwise) and the symbol [ begins a branch at an angle of 28.125 to the right (i.e. clockwise). The examples are all scaled to make their heights approximately equal; the relative scaling ranges from 1.0 for FIB to 0.026 for SCRUB. In the rst example (FIB) 716 separate line segments radiate from the starting point, many overlaid on each other. The resulting gure is 2 line segments tall. In the second example (SMITH, borrowed from [SMITH84]) 665 line segments branch from each other and produce a gure that is approximately 60 line segments tall. The replacement rules for SMITH are more complicated than those for FIB (although still quite simple), but even when comparing strings of approximately equal length from the two grammars, the graphic interpretation of SMITH is qualitatively more complicated and natural looking than that of FIB. Although changing the interpretation scheme could elicit a more ornate graphic for FIB, (e.g. by avoiding the overlaid segments), the extreme simplicity of the underlying structure makes it impossible to achieve the kind of complexity that we expect in a natural object. However, adding a little more variety in the structure (as in SMITH) appears to be sufcient to generate that complexity.
Figure V - a musical interpretation of the grammar from gure III The music in gure V is an interpretation of the seventh generation of FIB. The interpretation algorithm used for this example performed a depth-rst traversal of the tree. At each left bracketing symbol {[,{,(} a variable branch was incremented. At each right bracketing symbol {],},)} branch was decremented. At each alphabet symbol {a,b} a variable seg was incremented if the symbol was b, branch and seg were then used to select one of 50 sets of four notes, and branch was used to select one of 13 metric patterns that play some or all of the notes. This is one of the more complicated algorithms that were used. The fragment in gure V is pleasing and demonstrates a reasonable amount of musical variety. This contrasts sharply with the undeniably boring graphic in gure IV. We could draw any one of several conclusions from this discrepancy: The interpretation algorithm is insensitive to its input While an algorithm that simply ignores its input and emits the Moonlight Sonata (an extreme example) could obviously be written and would meet our criterion for musicality, it would sacrice the graph grammar benets of interesting structure (since we would be ignoring it) and database amplication (since the program would have to contain its entire output and would thereby be more complicated than the music it

generates). A reasonable test is to determine how much effect a change to the input has on the output. Obviously there is a matter of degree to consider here; some similarity in the outputs is to be expected (with human composers it might be called style), but we also expect easily recognized differences. When this algorithm is applied to other grammars, the output generated has recognizable similarities and differences, so we must conclude it is not insensitive to its input. Music requires less complexity than graphics. Music clearly requires different structure characteristics than graphics; the roles of ne and gross structure are almost exactly reversed in them. Music is sequential; it is experienced as a one-dimensional phenomenon that only varies with time (ignoring binaural effects), whereas (static) graphics is holistic; it is experienced as a two-dimensional phenomenon that does not vary with time. This is an important structural difference. When exposed to a graphic work, the rst impression the viewer receives is of the gross struc6 ture; further examination lls in the ner structure in a sequence determined by the viewers eye motion. When exposed to a musical work, the rst impression is of ne structure in one specic area (the rst measure), followed by ne structure in the next area (the next measure), and so on. Only after perceiving the ne structure can the listener know the gross structure of a musical piece. In his paper The Complexity of Song [KNUTH77], Donald Knuth proves the theorem There exist arbitrarily long songs of complexity O(1) by showing that a song recorded by Casey and the Sunshine Band has a lyric consisting of eight words arranged in a fourteen word long pattern and repeated arbitrarily many times.7 Although Knuths paper dealt with the text of songs (and was published as a joke), he touches on an important aspect of melodic structure; reiteration of previous segments. It is commonly accepted that expectation, based on the inductive/deductive process of guessing the overall structure and predicting what will come next, must be satised most, but not all, of the time in order for music to be pleasing. This can only be a consideration in a medium in which the structure is revealed sequentially. A simple way to achieve the middle ground between boring predictability and frustrating arbitrariness is to introduce a pattern and then repeat it with variations. The common implication is that high music tends away from predictability and low music (as implied in Knuths example) tends toward it. As one extreme, a long series of notes chosen randomly does not pass the is it musical? test. (Some very modern composers seem not to have noticed; is it because they are aiming at a very high audience?) Extreme complexity in music is perceived as arbitrariness, i.e. an excessively complex structure is perceived as no structure at all. Similarly, a screen full of many randomly colored pixels appears to be without structure and would fail the is it a graphic? test. On the other hand, a small number of randomly chosen notes or randomly colored blotches is often considered musical or graphical. One possible explanation is that since many patterns will match a small part of a random sequence, the small sample of randomness is perceived as a small sample of a real (i.e. predictable) pattern. An interesting measure would be how big a random pattern can get before it is recognized as meaningless. A comparison of these measures for graphic and musical entities might help answer the related question, does music tolerate less complexity than graphics? The structure in FIB is, for some reason, more appropriate for music than for graphics. The requirement that a balance be maintained between the repetition of old material and the introduction of new material in music implies that structures that contain repeats or near-repeats of segments are particularly appropriate. Static graphics has no time dimension in which temporal repetition can occur, and spatial repetition, while not unknown, is not as important in graphics as temporal repetition is in music. Although the syntax of graphics does not require repetition, the syntax of plants (which is the semantics of gure IV) does require repetition. Simply stated, plants manifest a structure composed of varied repetitions, just as music typically does. Is this a case of art imitating nature?8 The structure in FIB, which is almost entirely repetition, satises one of the important requirements of music (and plants), but has little to offer for graphics per se.

Progressive picture transmission techniques take advantage of this observation. Gross details are sent rst, followed by successively ner details until the viewer is satised [KNOWLTON80] [FRANK80]. 7 Thats the way, uh huh, uh huh, I like it, uh huh, uh huh,. 8 For when there are no words it is very difcult to recognize the meaning of the harmony and rhythm, or to see that any worthy object is imitated by them. [PLATO55]
192 samples of music were produced by trying twelve different interpretation algorithms on the third generation strings of each of 16 different grammars. The samples ranged from 0.0625 bars long (one note lasting 0.15 seconds) to 46.75 bars long (about 2 minutes). A small group of evaluators listened in randomized order to every sample and rated them on a 0 to 9 scale; 0 for awful through 3 for almost musical and 6 for pleasing to 9 for wonderful. Of these 192 samples 89% rated above 3.9 Some algorithms performed very well and generated not only musical but pleasing results on the average. Only one of the algorithms (the rst one designed) averaged below 3. If that algorithm is discarded the success rate becomes 95%. Eedie plays several examples of melodies generated by L-system grammars in her demo, including one in which sequences derived from two different grammars are interpreted by the same algorithm and overlaid. The similarities resulting from the common interpretation scheme give the piece a fugal quality while the differences in the grammars cause the two melodies to diverge in the middle and (by luck) converge at the end. Music Generated by Stochastic Binary Subdivision DDM The metric character of western popular music exhibits an extremely strong tendency. It is binary. Whole notes are divided into half notes, quarter notes, eighths, sixteenths, etc. And, while a division into three sometimes happens, as in waltzes or triplets, thirds are almost never subdivided into thirds again in the way 10 that halves are halved again and again. Further, it is rare for notes started on small note boundaries to extend past larger note boundaries. More precisely, if we group the subdivisions of a measure into levels, such that the nth level contains all the subdivisions which are at odd multiples of 2n into the measure (e.g. level 3 consists of the temporal locations {1/8,3/8,5/8,7/8}), we nd that notes started on level n infrequently extend across a level n1 boundary and only rarely extend across a level n2 boundary. Rhythms, like melodies, must maintain a balance between the expected and the unexpected. As long as, most of the time, the primary stress is on the one beat with the secondary stress equally spaced between occurrences of the primary stress, and so forth, that balance is maintained. By making note length proportional to the level of subdivision of the note beginnings, emphasis is constantly returned to the primary stress cycle. Simple attempts at generating rhythms randomly fail because they ignore this requirement. More sophisticated attempts make special checks to avoid extending past the primary or the secondary stress, but the result is still erratic because we expect to hear music that keeps reiterating the simple binary subdivisions of the measure. divvy(ip, lo, hi) struct instr *ip; int lo, hi; { int mid = (lo + hi) >> 1; ip>pat[lo] = |; if ((rand() % 100) < ip>density && hi lo > ip>res) { divvy(ip, lo, mid); divvy(ip, mid, hi); } } Figure VI - Subroutine divvy from ddm.c The program ddm11 attempts to make musical rhythms by adhering to the observations made above, and making all other choices randomly. Figure VI is the heart of ddm. The structure instr contains, among other things, density - a probability that, at any level, subdivision to the next level will occur, res - the

The median value was 5.0, but I would have been happy had it been 3. Waltzes and triplets probably gain much of their effect from the feeling that a beat is missing from a pattern of four, or has been added to a pattern of two. 11 a.k.a. digital drum madness
shortest note that can be generated, i.e. the deepest level to which it can subdivide, and pat - a character 45:75: 2:0: 96:1 BD 52:75: 4:0: 96:1 SD 57:50: 8:0:120:1 HHC 51:50: 8:0: 64:1 RIM 48:67: 8:0: 80:1 TOM3 54:70: 8:0: 64:1 CLAP 55:75: 8:0: 64:1 COWB 59:50:16:0: 80:1 HHO 53:67:16:0: 72:1 TOM1 Figure VII - Typical Data File for DDM string in which the generated pattern of beats is stored. Figure VII shows a typical input le for the program. The rst column is the code that must be sent to a drum machine to make the intended drum sound; the second column is the percent chance that subdivision will occur at any level; the third column is the smallest subdivision allowed; the fourth column is the maximum duration of a note (0 for drum machines); the fth column is how loud the sound should be; and the sixth column is which MIDI channel (i.e. which drum machine) should play it. Any further information on the line is comment, in this case the instrument A2 #.#. BD E3 |..#..|..#.. SD |.|. HHC Eb3 |... RIM |.#.|..|.#.|.. TOM3 Gb3 |.|.|.|.#. CLAP |.|.|.|.|. COWB B3 |.|.|..|. HHO |.|.#.|.#.|.#.|.#. TOM1 A3 C3 G3 F3
Figure VIII - Sample DDM Output name. Figure VIII shows the output of ddm in two forms; the rst is one measure of debugging information showing the results of all subdivisions and the second is two bars of drum score showing the nal result. Note that only one instrument is played at any one time; | indicates a subdivision, and # indicates the drum to be played at that time (pound sign seemed appropriate, somehow). Precedence is given to the instruments listed earliest in the input le, thus the bass drum (BD) plays the down beat, even though all the instruments wanted to appear then; similarly, the low tom-tom (TOM3) plays on the next eighth note, having precedence over the open hi-hat (HHO). The drum score in gure VIII starts with the same measure and then goes on for another measure which is quite different although based on the same set of probabilities and precedences. If we let the lines in the ddm le describe notes and pitch differences instead of drum types, ddm can be used to generate melodies. This is a simple change in the implementation; the drums are already encoded as pitches (45 A2 bass drum, 52 E3 snare drum, etc.). The only extension that is needed is to encode pitch differences. This we do by dening any value of 12 or smaller to be a pitch difference (by doing so, we make C#0 the lowest note we can specify directly). By adding lines like 1:60:16:31:64:0 and -1:65:16:31:64:0, meaning go up one step and down one step respectively, rising and falling motions can be Scale 1,2,4,7,9 Limits 48,84 69:33: 8:12:64:0 A4 64:33: 8:12:64:0 E4 1:60:16:28:64:0 up -1:65:16:28:64:0 down 1:55:32: 4:64:0 up -1:50:32: 4:64:0 down Figure IX - DDM File for Scat included. Figure IX shows an eight line le used to generate melodies to be sung by a Dectalk speech synthesizer. The Scale and Limits lines constrain the program to stay within a particular musical

- 10 -

scale (a simple pentatonic) and within a certain range (C3, an octave below middle C, through C6, two octaves above middle C). By specifying fairly short notes in the fourth column, the chance of rests appear-
Figure X - Sample Scat Output ing (for the singer to breathe) is increased. Figure X is a sample of ddm ouput from the le in gure IX. The program scat.c converts the MIDI format output of ddm into a sequence of Dectalk commands that produces scat singing. The syllables are generated by combining a randomly chosen consonant phoneme with a randomly chosen vowel phoneme; the result is usually humorous (and occasionally vulgar). Eedie uses ddm twice during the telephone demo, once to compose a little scat sequence to sing for the caller and once at the end to compose a piece for bass, drums, piano, and clavinet called Starchastic X, where X is the process id of Eedies process. Although no formal testing has been done with ddm, informal testing, (e.g. Well, howd you like it?), always elicits compliments with no apparent doubts as to its musicality. Results & Summary The telephone demo, despite its high down time (due in part to hardware frailties and in part to its dependence on two largely unrelated sets of experimental software), has been a great success. Eedie and Eddie have received over a thousand calls and have given as many as 60 demos a day to callers from 50 different area codes. Although no formal announcements of their demo has ever been made (prior to this paper), word of mouth has brought calls from Belgium, Canada, England, Holland, and even Texas. Although enough experience has been gained with the music generation algorithms to draw some tentative conclusions, it should be stressed that this is still a work in progress. From a quick perusal of the further work section it should be obvious that many ideas and experiments have yet to be tried, and that, undoubtedly, these tentative results will be modied and rened. The riffology technique generates musical sounds; it is sufcient but gets boring with repetition. Greater rhythmic variety and some interesting overall structure are needed. L-system grammars generate strings that have at least one important musical characteristic varied repetition. It is not yet clear whether the relationship between their ne structure and their gross structure is a benet for music. It is clear that the database amplication property is useful in that an arbitrarily long piece can be generated. The majority of cases tried generate musical sounds, allowing us to call this technique sufcient. The binary subdivision technique works reliably. A tendency to produce fast, rising or falling sequences must be avoided by careful le specications, otherwise almost comic sequences result (not unlike Chico Marx at the piano). Given that such sequences are avoided, its results are musical, so it is sufcient.

- 11 -

With all three differing algorithms showing sufciency we must conclude that none of the three is necessary. The interesting possibility remains that the overlap (intersection) of the three algorithms would be sufcient by itself. Unfortunately, the intersection of the three sets of rules is so trivial {use diatonic pitches, use no note shorter than a sixty-fourth note}, that it is unlikely that it is sufcient. It is interesting to note that many of the pieces generated by these algorithms appear to have semantic content, (some seem to brood, some are happy and energetic, others bring Yuletide street scenes to mind). Since the algorithms themselves have no information about human emotions or the smell of chestnuts roasting on an open re, we must conclude that any semantically signicant constructions that occur are coincidental. Our semantic interpretation of these accidental occurrences, like seeing a man in the moon, testies to the ability of human consciousness to nd meaning in the meaningless. Further Work The work presented here just scratches the surface of a very complicated subject area. For every idea implemented during this project, ten were set aside for later consideration. Among the ideas not yet tried for the riffology technique are: Expand the repertoire of riffs. Allow metric variation within the riffs. Enforce breathing, i.e. limit the length of phrases of rapid notes, inserting rests in the same way a singer must. Provide a mechanism for generating large-scale structure, perhaps by using grammar-driven or binary subdivision algorithms. Among the ideas not yet tried for grammar based techniques are: Experiment with much more complex grammars. Establish controls by feeding the interpretation algorithms random input and constant input. Compare the output from the controls to that from the grammars. Interpret the strings as rhythmic entities rather than melodic. Interpret the strings in larger chunks instead of symbol by symbol; perhaps by treating pairs or triples of symbols as objects, or by treating entire branches as objects. Experiment with 1L-systems or other context-sensitive grammars. Dene analogues in the audio domain for the graphic interpretation of the strings such that the sounds produced are, in some non-trivial sense, equivalent to the graphics produced. Choose an audio interpretation of the strings and design some grammars specically to sound musical with that interpretation; does the graphic interpretation of those grammars communicate the same components (whatever they are) that made the sounds musical? Among the ideas not yet tried for binary subdivision techniques are: Allow multiple simultaneous instruments or notes (currently only one note is played at a time). Allow the events being selected to be at a higher level than notes (e.g. entire phrases or repeats of previous output). Allow the events being selected to be at a lower level than notes (e.g. slurring of notes or vibrato). Allow the events being selected to have more global scope (e.g. key changes or tempo shifts). Among the ideas not yet tried for demonstrating music over telephone lines are: Solicit listener evaluations through the touch tone pad. Allow callers to select types of music or specify other parameters that inuence the music generated. Provide stereo output (via two telephone lines). Charge money. Acknowledgements Several people have been particularly helpful with this project. Gareth Loy of UCSD did the initial work with connecting the Sun to MIDI instruments, wrote the original record and play programs, and proved it can be done. Michael Hawley of the Droid Works extended Gareths work, provided a exible programming environment on the Sun, and, best of all, gave me copies of all that. Brian Redman is responsible for

- 12 -

the entire BerBell project and often went out of his way to allow me to do things with the phone system that no civilized person would allow. Stu Feldman (my boss) has provided encouragement, equipment, and proofreading. He has yet to suggest I consider a career in sanitation engineering. Credit is also due to fellow musicians who were excited by the idea of computers improvising music and contributed riffs: Steve Cantor, Mike Cross, Marty Cutler, Charlie Keagle, David Levine, Lyle Mays, Pat Metheny, and Richie Shulberg. References CHOWNING73 FRANK80 John Chowning, The Synthesis of Complex Audio Spectra by Means of Frequency Modulation Journal of the Audio Engineering Society vol. 21, no. 7, pp. 526534 Amalie J. Frank, J. D. Daniels, and Diane R. Unangst, Progressive Image Transmission Using a Growth Geometry Coding Proceedings of the I.E.E.E., Special Issue on Digital Encoding of Graphics, vol. 68, no. 7, pp. 897909 (July, 1980) Lejaren Hiller, Music Composed with Computers A Historical Survey, The Computer and Music, Cornell University Press, pp. 4296 (1970) Ken Knowlton, Progressive Transmission of grey-scale & binary pictures by simple, efcient, and lossless encoding schemes Proceedings of the I.E.E.E., Special Issue on Digital Encoding of Graphics, vol. 68, no. 7, pp. 885896 (July, 1980) D. E. Knuth, The Complexity of Songs SIGACT News vol. 9, no. 2, pp. 1724 (1977) P. S. Langston, The Inuence of Unix on the Development of Two Video Games, EUUG Spring 85 Conference Procedings, (1985) D. Levine & P.S. Langston, ballblazer, (tm) Lucaslm Ltd., video game for the Atari 800, & Commodore 64 home computers, (1984)

HILLER70 KNOWLTON80

KNUTH77 LANGSTON85 LEVINE84
LINDENMAYER68 Aristid Lindenmayer, Mathematical Models for Cellular Interactions in Development, Parts I and II, Journal of Theoretical Biology 18, pp. 280-315 (1968) MIDI85 MORGAN73 PLATO55 REDMAN85 REDMAN86 SMITH84 MIDI 1.0 Detailed Specication, The International MIDI Association, 11857 Hartsook St., No. Hollywood, CA 91607, (818) 505-8964, (1985) S. P. Morgan, Minicomputers in Bell Laboratories Research, Bell Laboratories Record, vol. 5, no. 7 p 194201 (1973). Plato, Laws, Book II, (ca. 355 B.C) B. E. Redman, Who Answers Your Phone in the Information Age?, Usenix Summer 85 Conference Proceedings, (1985) B. E. Redman, BerBell, Bell Communications Research Technical Memorandum (in preparation) Alvy Ray Smith, Plants, Fractals, and Formal Languages Computer Graphics Proceedings of the Siggraph 84 Conference, vol. 18, no. 3, pp. 110 (July 1984).
APPENDIX The following list contains brief descriptions of programs used to manipulate MIDI data or for some other aspect of the telephone demo. 2332probe Used by demo2332 to check puds status adjust perform metric adjustment dynamically atox convert hexadecimal ascii to binary bars count & select bars of MIDI data bbriffs generate improvisation for Song of the Grid ccc convert ascii chord charts to MIDI data chart display midi data in PRN (piano roll notation)

- 13 -

ched chmap cntl da dack ddm decsquawk harm inst just keyvel kmap m2midi m2p marteau marteauslide mc mdemo mecho merge metro midi2m midimode mpuclean mudef muzak notedur p0la p0lb p0lc p0ld p0le p0lf p0lh p0li p0lpic play punk ra reclock record rxkey scat select stretch transpose trim tshift unjust xtoa
interactive midi data editor (CHart EDitor) select and remap MIDI channel events generate MIDI control change commands disassemble MIDI data to ASCII listing lint for ASCII MIDI listings (da check) stochastic binary subdivision generator control a Dectalk voice synthesizer harmonize melody lines (MIDI) generate MIDI voice change commands adjust (quantize) note timings (MIDI) scale, compress, and expand key velocity dynamics remap key values (pitches) convert m-format scores to MIDI convert m-format scores to PIC macros for typesetting generate and play modern (random) music interactive interface for marteau C compiler with MIDI libraries introduce, compose, and play the telephone demo add echo to MIDI data combine MIDI data streams metronome with graphic interface convert MIDI data to m-format scores remove running status from MIDI data streams insert running status, remove other junk from MIDI data PIC macros for typesetting music scores interpret unsuspecting ASCII text as music change note lengths without changing tempo a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a musical interpretation of 0L-systems a graphic interpretation of 0L-systems for typesetters play (and overdub) MIDI data streams generate a soft-punk melody and accompaniment assemble (re-assemble) ASCII data into MIDI regenerate timing commands for MIDI data record MIDI input print information about the Yamaha RX/11/15/21 key mappings convert MIDI data into vocal scat for the Dectalk lter MIDI data by various criteria retime MIDI data transpose MIDI key event (pitch) data remove dead air from MIDI data shift MIDI data temporally add slight random time variations to MIDI data convert binary data to hexadecimal ASCII