I don't quite know why it has taken me so long to get around to posting this, but my dissertation (from June, 1999) may be of interest to some of the members of this list. A one sentence description of the work (as relevant to this list) is "a computational model of musical-instrument recognition within a computational auditory scene analysis framework". Abstract, TOC, and URL below. I hope some of you find it useful and/or interesting.
Cheers,
--Keith
-----
Keith D. Martin
Bose Corporation
(formerly of the MIT Media Lab)
The Mountain (R&D - 15C)
Framingham, MA
01701-9168 USA
Relevant details:
Martin, Keith D. (1999) Sound-Source Recognition: A Theory and Computational Model. Ph.D. Thesis. Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science.
You can download a copy of the document here: ftp://sound.media.mit.edu/pub/Papers/kdm-phdthesis.pdf (2.3 MB).
Abstract
The ability of a normal human listener to recognize objects in the environment from only the sounds they produce is extraordinarily robust with regard to characteristics of the acoustic environment and of other competing sound sources. In contrast, computer systems designed to recognize sound sources function precariously, breaking down whenever the target sound is degraded by reverberation, noise, or competing sounds. Robust listening requires extensive contextual knowledge, but the potential contribution of sound-source recognition to the process of auditory scene analysis has largely been neglected by researchers building computational models of the scene analysis process.
This thesis proposes a theory of sound-source recognition, casting recognition as a process of gathering information to enable the listener to make inferences about objects in the environment or to predict their behavior. In order to explore the process, attention is restricted to isolated sounds produced by a small class of sound sources, the non-percussive orchestral musical instruments. Previous research on the perception and production of orchestral instrument sounds is reviewed from a vantage point based on the excitation and resonance structure of the sound-production process, revealing a set of perceptually salient acoustic features.
A computer model of the recognition process is developed that is capable of "listening" to a recording of a musical instrument and classifying the instrument as one of 25 possibilities. The model is based on current models of signal processing in the human auditory system. It explicitly extracts salient acoustic features and uses a novel improvisational taxonomic architecture (based on simple statistical pattern-recognition techniques) to classify the sound source. The performance of the model is compared directly to that of skilled human listeners, using both isolated musical tones and excerpts from compact disc recordings as test stimuli. The computer model's performance is robust with regard to the variations of reverberation and ambient noise (although not with regard to competing sound sources) in commercial compact disc recordings, and the system performs better than three out of fourteen skilled human listeners on a forced-choice classification task.
This work has implications for research in musical timbre, automatic media annotation, human talker identification, and computational auditory scene analysis.
Table of Contents
1 Introduction
1.1 Motivation and approach
1.2 A theory of sound-source recognition
1.3 Applications
1.4 Overview and scope
2 Recognizing sound sources
2.1 Understanding auditory scenes
2.1.1 Exploiting environmental constraints
2.1.2 The importance of knowledge
2.1.3 Computational auditory scene analysis
2.2 Evaluating sound-source recognition systems
2.3 Human sound-source recognition
2.4 Machine sound-source recognition
2.4.1 Recognition within micro-domains
2.4.2 Recognition of broad sound classes
2.4.3 Recognition of human talkers
2.4.4 Recognition of environmental sounds
2.4.5 Recognition of musical instruments
2.5 Conclusions and challenges for the future
3 Recognizing musical instruments
3.1 Human recognition abilities
3.2 Musical instrument sound: acoustics and perception
3.2.1 An aside on "timbre"
3.2.2 The magnitude spectrum
3.2.3 The dimensions of sound
3.2.4 Resonances
3.3 Instrument families
3.3.1 The brass instruments
3.3.2 The string instruments
3.3.3 The woodwind instruments
3.4 Summary
4 Representation
4.1 Overview
4.1.1 Mid-level representation
4.1.2 Features and classification
4.2 The front end
4.2.1 Bandpass filterbank
4.2.2 Inner hair cell transduction
4.2.3 Pitch analysis
4.3 The weft
4.4 Note properties / source models
4.4.1 Spectral features
4.4.2 Pitch, vibrato, and tremolo features
4.4.3 Attack transient properties
4.5 The model hierarchy
5 Recognition
5.1 Overview and goals
5.2 Definitions and basic principles
5.3 Taxonomic classification
5.3.1 Extension #1: Context-dependent feature selection
5.3.2 Extension #2: Rule-one-out
5.3.3 Extension #3: Beam search
5.4 Strengths of the approach
5.5 An example of the recognition process
6 Evaluation
6.1 A database of solo orchestral instrument recordings
6.2 Testing human abilities
6.2.1 Experimental method
6.2.2 Results
6.2.3 Discussion
6.3 Computer experiment #1: Isolated tone pilot study
6.4 Computer experiment #2: 6- to 8-way classification
6.5 Computer experiment #3: Direct comparison to human abilities
6.6 General discussion
7 Summary and conclusions
7.1 Summary
7.2 Future developments
7.3 Insights gained
7.4 Conclusions
References