ISE599 - TEAM-BRE Project

TEAM-BRE Project
	Team Members: Ananda Jacobs Beran Pacaci

Introduction
Musicians frequently play synthesizers, samplers and sequencers in today's music. Such music can be created also by computers. But they result in a dull, mechanical performance. A better result can be obtained by applying a generative performance grammar to the original music piece. This study aims to show that timbre, or characteristic sound of an instrument, plays a role in how listeners perceive expression in music.
Project Description
The Team-bre project aims to implement the expressive reproduction capabilities of the Director Musices program to explore different timbres. The theory behind the project is that timbre plays an important role in expressivity. Players of different instruments were hypothesized to employ different types of expression rules. To demonstrate, two cello players, two clarinet players, and two trumpet players were asked to play a Chopin Etude (Opus 10, No. 3). Director Musices was then used to try and approximate these performances. The aim was to generate instrument-specific rule sets. The rule parameters for one instrument (example, cello) were then applied to the nominal MIDI versions of other instruments. We hypothesize that listeners will prefer the rules matched with the original instrument, rather than a different instrument. Thus, cello rules should sound best on cello.
Experiment Details
We started our experiment by recording music performances that were played with different instruments. We asked musicians from the Thornton School to play a Chopin Etude (Opus 10, No. 3). We took the best out of two performances from each player. None of them had seen the music before the experiment took place. A mini disc (Sony MD Walkman, MZ-R90) and a regular microphone (Sony Microphone, ECM-717) were used for recording. Then the files were transferred to the mp3 format using Cool Edit 2000 tool. A noise reduction filter was applied to get a clear sound quality. We made a nominal file for each of the four instrument types. A nominal file is a MIDI file with no expressivity at all –just the notes at a steady tempo. This was done with Reason 2.5 and a MIDI keyboard, with the input tool set to “tempo lock.” This was to ensure that the note values were snapped to a grid, minus any expressivity. In order to get a MIDI version of the performances, we had to approximate them using a MIDI keyboard and Reason, because we tried in IntelliScore (Polyphonic MP3 and WAV to MIDI Converter v5.1) program to convert mp3's to MIDI's but the results were not satisfactory. While the actual mp3 performance played in headphones, the “fake” performer (Ananda) copied the performances. While inherently biased, this was the only way to get a reasonable transfer from mp3 to MIDI. The MIDI files were assigned general MIDI timbres for piano, cello, clarinet, and trumpet. In Director Musices: We changed the tempos if necessary, to accommodate players playing the piece faster or slower. To set the instrument rules, we picked the best performance for each instrument. The parameters we used were High-Loud, Duration Contrast (2 types), Punctuation, and Final Ritard. With the exception of High-Loud, all of the parameters dealt with alterations of note duration. We did not use Phrase parameters because there were no phrase markings in the score. First of all, we applied the following rules one by one. You can find the definitions of the rules and the results of the experiment on the links below: High-Loud Duration-Contrast Punctuation Final Ritard Our next step was to apply a combination of rules for each instrument in order to approximate their performances in DM. Director Musices performance approximations: Music Performance Approximation for Piano Music Performance Approximation for Cello Music Performance Approximation for Clarinet Music Performance Approximation for Trumpet Our last step was to apply the combination of rules that are chosen for a particular instrument performance to the other instrument performances. Cello Performance Approximation Rule Set applied to Other Instrument Performances
Difficulties in Approximation
Some performance parameters were difficult to approximate using the DM system. slight pause before pitch peaks speeding up with sequences getting higher slowing down before cadences other than at end pauses between “phrases” that were imposed by performers fluctuating tempos and other parameters that were not rule-based or regular exaggerated pause before very last note Some things in performances are spontaneous or unintentional. For example, First cellist played the entire second half wrong, thinking it was a repeat of the first half. Second clarinetist played wrong note in first half. Second trumpeter articulated when the score dictated a slur. First cellist played slightly faster what had already been played.
Notes
These were generally spontaneous performances. Perhaps if the musicians had further study of the piece, they would change the “expressivity” of the performance. We were unable to output graphic correlations between real performance and approximated performance. To the best of Team-bre's judgement, the approximations sounded as close as they were able to get to the original file. Since we had to approximate MIDI performances, the expressivity was undoubtedly skewed. The High-Loud rule was probably especially inaccurate, since volume control on piano, cello, clarinet, and trumpet are very different from volume control on a MIDI keyboard. It is important to note that individual instruments had very different timbres. Comparing the mp3s of players 1 and 2 for each instrument will show this. For instance, one trumpet sounds very "warm" and "dark," while the other trumpet is "light" and "thin" sounding. The general MIDI sound database only gives one option for each instrument, which is not enough to cover individual differences.
Conclusion
Instrument-specific rules Piano: The duration contrast made the piano sound too choppy and unnatural. Punctuation was somewhat good at capturing articulation, but less so than for the clarinet and trumpet. Cello: Relatively little High-Loud rule was used. One explanation is that bowing generally uses even pressure on the strings, making for less dynamic contrast. In fact, the dynamic range is smaller on the cello than it is on the clarinet. Less punctuation and less ritardando were also characteristic of cello performance. One explanation is that slowing down is more unnatural for string players because it means the player will have a more difficult time maintaining volume. Articulation tends to be added more often in clarinet and trumpet, whereas itʼs only done if specified in the score for strings. There were no specific markings for articulation, so the default for cello is to play everything legato. (This assumption is not made for clarinet.) The cello tends to accelerate to a pitch peak, then decelerate when coming down off a peak. Clarinet: More Punctuation was needed, because clarinet players have to stop more often to take a breath. Unlike the cello and piano, who can continuously play without stopping, the clarinet must insert breaks from time to time. More High-Loud was used in the clarinet. This is likely because the clarinet is physically a more dynamically capable instrument than the cello, for instance. In every range, the volume can be very very soft or very very loud. So, players tend to make use of this capability. Trumpet: The particular trumpet players took more time to play the Etude, so even more breathing was necessary. The breaks tended to be especially exaggerated when the melody went lower.
Software
Several applications are used throughout the project for different needs. Director Musices v2.5 Cool Edit 2000 GUIDO NoteViewer v0.9 Reason v2.5
References
Johan Sundberg, Anders Friberg and Roberto Bresin (2003). Attempts to Reproduce a Pianist's Expressive Timing with Director Musices Performance Rules. The Science of Music Performance. ( http://www.speech.kth.se/music/performance )
Last Update: April 29, 2004