Outline

The treatment of speech signal processing requires an initial grounding in digital signal processing. This not only enables students to understand why things appear the way they do in various spectrographic representations, but also allows a proper coverage of basic speech processing algorithms such as linear prediction and cepstral analysis. With these first principles, students will be able to anticipate the problems with and follow the motivation for nearly all commonly-used speech processing methods. The actual speech processing algorithms to be covered are not prescribed. Where possible, students should have access to a software environment which allows interactive investigation of the basic algorithms.

Topics:

• Signal processing tools
• digital filters
• Fourier series and transforms, DFT, FFT
• Short-Term Fourier Transform (STFT)
• Filter banks
• Speech acquisition and digitisation
• Speech analysis and parameter extraction
• Short-term analysis, frames and windows
• Time-domain analysis: energy, zero-crossings, statistic parameters, autocorrelation
• Frequency-domain analysis: spectra and spectrograms
• Cepstral analysis
• Linear prediction analysis
• Pitch and formant estimation
• Static and dynamic features
• Speech signal synthesis
• Speech coding
• Speech enhancement

References

1. Owens, F.J. (1993), Signal Processing of Speech, Macmillan.

2. Deller, J.R., Proakis, J.G., and Hanson, J.H. (1993), Discrete-Time Processing of Speech Signals, Macmillan.