Naked Science Forum

Non Life Sciences => Geek Speak => Topic started by: syhprum on 18/01/2012 11:28:36

Title: What is the minimum bitrate for intelligible speech transmission?
Post by: syhprum on 18/01/2012 11:28:36
Has any one experience with vocoder encoding/decoding software I understand quite intelegable speech can be sent at 600b/s and would like to make tests as to what cable bandwidth would be required to carry this.
Title: Re: Vocoder encoding
Post by: RD on 18/01/2012 12:30:20
A free program called SOX (https://en.wikipedia.org/wiki/SoX) can convert audio into LPC-10 (https://en.wikipedia.org/wiki/FS-1015), which has a very low bit rate (but sounds like "speak n spell (https://en.wikipedia.org/wiki/Speak_%26_Spell_%28toy%29#Electronics)").

Below is an example of variable bit rate LPC-10, varying from 356 bps - 2.4 Kbps, averaged 1905 bps 
http://hawksoft.com/hawkvoice/samples/vbrlpc10.wav


[ 800b/s is the lowest I can find ... http://asadl.org/jasa/resource/1/jasman/v103/i5/p2778_s1?bypassSSO=1 ]

Title: Re: Vocoder encoding
Post by: graham.d on 12/02/2012 09:16:10
Yes Syphrum, I have been involved with design and manufacture of vocoders and given a paper on the subject. If you just want speech (but sounding like Stephen Hawking) you can get well below 600Hz. Usually the requirement is to have other aspects of voice quality so that you could, for example, recognise the voice of the person speaking and, maybe, have options to reject background noise or, alternatively, make sense of it. Typically a minimum with current technology is 2400 bits/second which get "near Toll Quality" meaning nearly as good as a good telephone line with a speech bandwidth up to about 3500Hz. If you want to send this some distance by wireless or cable you would need to have some Forward Error Correction bits added - say making the overall bitrate 3600b/s. All these vocoders use some sort of linear predictive algorithms and utilise knowledge of how the vocal tract works and the brain interprets the received sounds. Don't expect them to make any sense of music though! They are widely used by the military and para-military (police/fire/ambulance) in digital VHF radios which are required to use narrow bandwidths. Some vocoder software/hardware is designed to be configurable to bandwidths as low as 600Hz (MELPe) but as you drop the bandwidth the sound becomes increasingly mechanical sounding though still intelligible. Popular commercial vocoders include AMBE, RALCWI and TWELP but they are not free as software. You can buy chips with some of these Vocoders on them though, along with suitable ADCs and DACs.