LPC for speech (fwd)
<jp12745@coewl.cen.uiuc.edu>jerry@terminus.us.dell.com> wrote:
Did you know that using LPC (linear predictive coding) on speech can near-telephone quality at only 8 k BITs/second? With a signficant decrease in quality (but still very understandable... probably better than radio) you can get the rate down to 2kbps. If you don't mind sounding like a speak&spell, you can go to 600bps or less.
haha... Geez oh man I just had to laugh at that speak&spell comment... Yeah, those things had funky voices, but it was pretty darn good speech synthesis for the time. (around 1982-3? I guess) (It could say alot more than my Apple IIe :) But those things had software with little bugs (or undocumented features) if you pressed combinations of keys at the same time it would do weird things. I kinda wish I still had some of those things, so I could try reprogramming it to do interesting stuff. (Imagines scene in US customs office: "Crypto exports? No, sir, this isn't a secret cipher machine, just an old speak&spell! Honest! You didn't think I was going to try to phone home with it or something did you?" ;) but I digress... 600bps is not realistic. Most people can READ text faster than 600 bps (okay it's a little above average, but not if you were skimming; to give you a rough estimate, it's a little slower than reading one line per second) You just can't expect to cram all the intricacies and inflections of speech into a 600bps channel and be understood. But why sacrifice quality when you can use a 14400bps modem? They aren't all too expensive these days.
Using LPC, you could send real-time voice over the internet. It would even work (maybe just barely) over a SLIP connection. According to my professor, LPC can be implemented in a simple DSP chip, so I figure a 486 ought to be able to handle it, too. Sound like an interesting (granted maybe not too useful) project? It would be a way of providing secure voice communications -- LPC code the speech, encrypt the data stream, transmit via v.32bis modem, etc.
Not too useful??? Sounds very useful to me. If someone got this working, it would be all we'd need to kill the clipper chip completely. Who would spend several hundred dollars for a "secure" phone when you can do the same (or better) with a $50 sound card for your computer? (Well, some might need to upgrade their modems too, but it would be a good idea to do that anyway.) Home (or office) computer cryptophone = no need for clipper. loki@cass156.ucsd.edu wrote in response:
I am working on this very thing. We will be using LPC encoding for compression, IDEA for encryption, and DH key exchange for key handling. We plan to use something better than DH ASAP (something less vulnerable to man in the middle attacks). We plan to use 14.4kbps transmission speed.
What kind of hardware will you be using? 486? DSP? or something 68xxx-based? What additional hardware (sound cards) will be supported/required? Will source be availiable?
<jp12745@coewl.cen.uiuc.edu>jerry@terminus.us.dell.com> wrote:
Did you know that using LPC (linear predictive coding) on speech can near-telephone quality at only 8 k BITs/second? With a signficant decrease in quality (but still very understandable... probably better than radio) you can get the rate down to 2kbps. If you don't mind sounding like a speak&spell, you can go to 600bps or less.
[...]
600bps is not realistic. Most people can READ text faster than 600 bps (okay it's a little above average, but not if you were skimming; to give you a rough estimate, it's a little slower than reading one line per second) You just can't expect to cram all the intricacies and inflections of speech into a 600bps channel and be understood. But why sacrifice quality when you can use a 14400bps modem? They aren't all too expensive these days.
600 bps is not realistic? Voice has been gotten down to 400 to 600bps by doing some coding on the output of a LPC (pitch + gain + filter parameters) and transmitting that. The remote decodes to recover LPC parameters and uses that to synth speech. The quality is supposed to be the same as 2400 bps LPC (slightly synthetic sounding, definitely not 'toll quality').
participants (3)
-
jerry@terminus.dell.com -
Matthew J Ghio -
Timothy Newsham