Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!ames!ll-xn!tj From: tj@XN.LL.MIT.EDU (Thomas E. Jones) Newsgroups: sci.electronics Subject: Re: RS-232 for voice Summary: Yes, but be more specific. Keywords: LPC sinusoudal transform segment vocoder Message-ID: <1490@xn.LL.MIT.EDU> Date: 27 Jul 89 13:30:42 GMT References: <8616@cbnews.ATT.COM> Reply-To: tj@xn.UUCP (Thomas E. Jones) Distribution: usa Organization: MIT Lincoln Laboratory, Lexington, MA Lines: 49 Can voice be transmitted over RS-232 lines? The answer is yes, and it is, but you must be more specific. If you want to speak in a microphone, into a computer, and have speech sent serially there are many ways to do it, but it is fairly complex. Data rates of understandable speech can vary from 9600 baud to 300 baud by using different methods. I'll tell about these below. If you don't want real-time vocoding, you just want to transmit data representing speech, you can send the ASCII characters down the line, and then have one of radio-shacks text-to-speech chips on the other computer, to convert to speech. Alternately, you can break the words you want to send into phonetic pieces, and send codes for these pieces down the line, at an even lower rate than the ASCII spellings of the words. Toys such as Speak & Spell store phonetic data like this in ROMS. Parts cost is around $20, if you already have the computers to transmit serial data. If you want to run real-time vocoding, (speak into microphone, have computer process it, and send it over serial lines.) there are many ways to do it. You can use LPC vocoding, which is in common use in the military. GTE makes several miltary vocoders to do this at baud rates around 9600, or 4600 baud, but it sounds bad at these lower rates. I think 9600 baud is about the worse you can deal with. You can code this LPC algorithm on a DSP microprocessor for around $100 in parts. Look in the library under LPC coding of Speech. MIT Lincoln Laboratory recently developed an algorithm called "sinusoidal coding of speech." It sounds much better at lower baud rates. It makes LPC coding sound terrible in comparison. 4800 bps data rates sound very natural. I can give you many references on this (call or email.) BBN recently developed a very low rate vocoding algorith, called the "LPC Segment Vocoder" that was recently (what time is it?) implemented in real-time by MIT Lincoln Laboratory. This codes speech at 300 bps, and sounds fairly understandable. It relies on tremendous lookup tables of several megabytes, and requires lots of hardware. I can also give many references on this algorithm (call or email.) DISCLAIMER: I am no expert in this area, but worked for several people experts for a period of time. All information is from the top of my head, and could be wrong. My employer doesn't necessary share in any of my opinions. - Thomas E. Jones tj@xn.ll.mit.edu (617) 981-5093 (work) -- tj@xn.ll.mit.edu or tj@ll-xn.arpa (one of these should work) Thomas E. Jones, home (617) 279-0767 work (617) 981-5093