[comp.dsp] 300 Baud Speech Vocoding...

tj@xn.ll.mit.edu (Thomas E. Jones) (11/16/90)

I'm a little surprised no one has talked about the low data-rate
speech vocoding algorithms that have been developed.

Here at the laboratory, we implemented a real-time 300 Baud algorithm
developed at BBN called an LPC segment vocoder.  Yes, it takes a lot
of computation to do it in real time, and a lot of computer hardware,
but shows that low data rates are possible, and there remains lots of
work to be done on improving these.

A new vocoding algorithm known as Sinusoidal Coding has been
developed by the same group at Lincoln Lab., and offers much better voice
quality than LPC.

I'm no expert in this area, but can point people toward references if
anyone's interested.

				- tj@xn.ll.mit.edu
-- 
tj@xn.ll.mit.edu or tj@ll-xn.arpa          (one of these should work)
Thomas E. Jones, home (617) 924-8326 work (617) 981-5093

wilf@sce.carleton.ca (Wilf Leblanc) (11/16/90)

tj@xn.ll.mit.edu (Thomas E. Jones) writes:

>I'm a little surprised no one has talked about the low data-rate
>speech vocoding algorithms that have been developed.

Me too. I thought with all the net bandwidth regarding compression
to 'walkman' quality, someone would have brought this up.  Maybe
not walkman quality, but at least it is 'intelligable' (at a
very low data rate).

>[stuff deleted regarding segment vocoder @ 300 bps]

There is alot of literature regarding 'vocoders' at 300 (or so) bps.
Basically, most of them are an LPC vocoder (LPC10 type) with
Vector Quantization of the LPC parameters and so on.  See ICASSP
around 1980-1984.  Not (as) much written on it since then (to my
knowledge).

>A new vocoding algorithm known as Sinusoidal Coding has been
>developed by the same group at Lincoln Lab., and offers much better voice
>quality than LPC.

Hmmmmm.  Correct me if I'm wrong, but I thought Sinusoidal (Transform)
coding was a (relatively) old technique which is becoming more popular
due (mostly, if not entirely) to the work of McAuley and Quateri (my
apologies if their names are spelled wrong, but this is on memory).


>[.....]

>				- tj@xn.ll.mit.edu
>-- 
>tj@xn.ll.mit.edu or tj@ll-xn.arpa          (one of these should work)
>Thomas E. Jones, home (617) 924-8326 work (617) 981-5093

--
---
Wilf LeBlanc                                Systems and Computer Eng.
Internet: wilf@sce.carleton.ca              Carleton University
    UUCP: ...!uunet!mitel!sce!wilf          Ottawa, Ont, Canada