The Speech Processing Cookbook
------------------------------

The aim of this exercise is to produce set of routines in C and Matlab
to execute all the common speech processing algorithms.

The aim is not state of the art recognition, synthesis or coding, but to
provide an introduction to these subjects, so providing breadth, not
depth.

The current status of this proejct is pre-alpha release.  The SLICENSE
file gives the conditions on copying and useing this software - these
may well change in the future.

Appended is a guess at what the table of contents might look like when
finished.

If you would like to contribute (library code, complete example
programs, documentation, bug fixes,... ) then please get in touch.


Tony Robinson

------------------------------------------------------------------------------
Dr AJ Robinson Phone +44 223 332815 Fax +44 223 332662 Email ajr@eng.cam.ac.uk
Cambridge University Engineering Department, Trumpington Street, Cambridge, UK
------------------------------------------------------------------------------

The Speech Processing Cookbook
------------------------------

Intro (brief)
	The physiology of speech generation and perception
	Waveforms
	Acoustics 
	Phonetics
	Linguistics	
	
Basics
	coding style, arg passing, arrays
	sampling
	filtering
		FIR
		IIR
		sample rate convertion
	Windowing
		hamming
		sizes
		zero padding

Acoustic Features
	Short term energy
		Broadband signals: RMS
		Narrowband signals: Hilbert transforms		
	Frequency scales
		Mel, Bark, ERB,...
	Filter banks
		Crude cochlear models
	FFT based
		FFT code
		real FFT
		cepstra
		Mel cepstra
	LPC based
		autocorrelation (direct and via FFT)
		autocorrelation to reflection and lpc coefficents
		log area ratios
		line spectral pairs
		linear LPC cepstra
		bilinear LPC cepstra
	Pitch
		time domain (zero crossings etc)
		frequency domain (autocorrelation/celstral)
		[ demonstration pitch tracking ]
	Formants
		peak picking
		root solving
		continuity constraints
	Voicing
		autocorrelation
	Speed changes
		segment based
		phase vocoder

Coding
	Coding distributions
		General (Huffman/arithmetic)
		Special	cases
	Vector Quantisation
	Sample coding
		no prediciton: ulaw/alaw
		Linear prediciton: ADPCM
	LPC coding
		Synthesis from LPC coefficents
		Mixed excitations
		Quantising LPC coefficents
		LPC10e
	Other schemes
		GSM
		CELP
Synthesis
	parametric
		Formant based
		smoothing
	diphone
		dicitonary generation
		smoothing
	Higher level considerations
		pronunciations
		syntax
		prosody			
Recognition
	Relevant acoustic features
	Crude linear match
	Dynamic Time Warping
	Markov Models
		Hidden (VQ, Gaussian Mixtures)
		Direct (Neural Nets)

-----------------

Copies shipped so far:

egrep cookbook ~/xferlog | sed s'/^.*\(cookbook[^ ]*\).*$/\1/' | uniq -c

      2 cookbook-01Feb94.tar.Z
      2 cookbook-03Feb94.tar.Z
     86 cookbook-04Feb94.tar.Z
      3 cookbook-06Apr94.tar.Z
     69 cookbook-08Apr94.tar.Z
      3 cookbook-25May94.tar.Z
     28 cookbook-26May94.tar.Z
     82 cookbook-03Jun94.tar.Z
	cookbook-23Jun94.tar.Z

275 copies in 142 days