Linux Phone Project

Digital voice using 4,800 bps modems...

NEWS UPDATE November 22, 1996
There have been 321 successful downloads of the source code since November 15, 1996.

There's now a mailing list, see below.

Features:

Real Time Linear Predictive Coding (LPC) coder
Compression better than an astounding 97%
Compress either sound files from disk or real time (from /dev/dsp)
File player
Network support allows voice link (half-duplex at present) using standard slow-speed dial-up modems (4,800 bps or better)
Multicast daemon (server process) for audio distribution to several clients simultaneously

Requirements:

Linux (and other Un*x systems, untested)
Sound Card and driver supporting /dev/dsp interface
(Optional) For network support, you need a TCP/IP Net connection (4,800 bps or better)

Future Possibilities: (This does not represent a commitment)

Full Duplex voice operation
Encryption for secure communication support
Audio Multiplexing
Integration with the Web via CGI (server) and plug-ins (client)
Voice Mail

Availability: (Help Wanted: Alpha Testers)
Click here to download source code
Click here to download a sample .WAV file
Send comments via e-mail to Max Parke

Copyright:
Free Software (GNU Public License)
Includes public domain LPC software developed by the United States Government

MAILING LIST:
The mailing list's E-mail address is lpp-l@lightlink.com.

If you wish to join the mailing list (open to all parties interested in further development of this project), send E-mail to majordomo@lightlink.com.
You may leave the `Subject:' field blank.
In the BODY of the message:
subscribe lpp-l

To unsubscribe, send E-mail to majordomo@lightlink.com.
You may leave the `Subject:' field blank.
In the BODY of the message:
unsubscribe lpp-l

Further Information
The speech coder uses approx. 25% CPU of a P5-100 for real time operation. The decoder uses about 12-13%. The encoder can read WAV format files or for real-time sampling reads /dev/dsp. The decoder can output audio either to the speakers or to a file. The system uses 16 bit mono data sampled at 8,000 per sec.

There is also a network option allowing the encoder and player to run on different machines. You can also have multicast servers which use close to zero CPU time when serving pre-coded files.

See Also
There is at least one other similar facility available, called Nautilus

Audio Samples
This demo file was captured from one channel of a stereo FM broadcast and encoded at 8,000 sample/sec, 16 bit. The file starts with the voice of a male announcer (Daniel Shorr) and ends abruptly during a weather broadcast by a female announcer. During a portion of the sample there is some light jazz guitar. The compressed version of this file is 53,248 bytes!

The sample sound file can be downloaded here.

Bugs Fixed
Version 0.2 is now available, with fixes for two bugs. First, the sound support would only allow `/dev/dsp', but not `/dev/dsp1'. Second, if your sound card's DSP buffer size was less than 4,096 bytes, the program would refuse to start. This check has been removed, so now any DSP buffer size is permitted. The buffer size is read from the sound card at program startup time.

Hopefully these have been fixed, please let me know!