Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!doc.ic.ac.uk!uknet!cam-eng!dsl!ajr
From: ajr@dsl.eng.cam.ac.uk (Tony Robinson)
Subject: Re: Time Encoded Speech
Sender: news@eng.cam.ac.uk (Usenet News)
Message-ID: <AJR.93Aug20171426@dsl.eng.cam.ac.uk>
In-Reply-To: ee91ajh@brunel.ac.uk's message of Fri, 20 Aug 1993 14: 38:25 GMT
Date: Fri, 20 Aug 1993 16:14:26 GMT
References: <CC2BC2.93I@brunel.ac.uk>
Nntp-Posting-Host: dsl.eng.cam.ac.uk
Organization: Engineering Department, Cambridge University, England.
Lines: 33

In article <CC2BC2.93I@brunel.ac.uk> ee91ajh@brunel.ac.uk (Adam H) writes:
>
> Does anybody know where I can get hold of algorithms/papers/source code
> for TES (Time Encoded Speech) systems?

The original reference is:

	R. A. King and W. Gosling, "Time-Encoded Speech", Electronics
	Letters, 20th July 1978, Vol 15, no 15.

The abstract says:

	A new method of digitising speech waveforms is described, based
	in the comparison of sucessive segments of the waveform with a
	suitably stored catalogue of possible distinct shapes.

The paper talks about bit rates of 5400bps to 3600bps.  Whilst this is
not stunning by todays standards, this quick and dirty technique has a
lot of appeal for real time systems without dedicated hardware support.

The algorithm is simple.  Take a speech waveform and segment it on the
zero crossings.  Make a code book of these half wave sections using
vector quantisation.  Now, for any new speech, just transmit the code
book index.  The code book can be crude, for example half sine waves
works quite well.  The codebook search can then be made trivial by
finding the power (or sum of absolute values) in the half wave,
transmitting this and the number of samples between zero crossings, and
reconstructing with the appropriate half sine wave.

All good fun.


Tony Robinson
