izumi@violet.berkeley.edu (Izumi Ohzawa) (12/10/88)
In article <21949@apple.Apple.COM> desnoyer@Apple.COM (Peter Desnoyers) writes: >In article <19355@ames.arc.nasa.gov> bhine@nike.UUCP (Butler Hine sst) writes: >>I kept the sample of NeXT voice mail sent out a while ago, but I never >>saw any explanation of the format. Does anyone know what the format is? > >I posted that message. The format is simple, like uuencoded data but >just a hair different. The binary data is standard mu-law PCM voice at >8k samples/second, with 8 bit samples. > Peter Desnoyers Are you sure that this is the format NeXT voice mail is transmitted over the net? I grabbed that file, and it was HUGE. At 8 kbytes/second of voice, I am not sure if anyone can tolerate the traffic. A lot of foreign e-mail addressees pay $$$ per kilobyte of messages RECEIVED and transmitted. I think Japanese users pay something like $1.00/kilobyte. That comes to $8/second! Also, with straight PCM, the much touted DSP chip isn't necessary. Surely, something like LPC (linear prediction coding) or PARCOR (I forgot what that stands for) can be implemented easily with the DSP chip. Izumi Ohzawa izumi@violet.berkeley.edu ...ucbvax!violet!izumi
desnoyer@Apple.COM (Peter Desnoyers) (12/13/88)
In article <18109@agate.BERKELEY.EDU> izumi@violet.berkeley.edu (Izumi Ohzawa) writes: > >Are you sure that this is the format NeXT voice mail is transmitted >over the net? I am sure that the file which I grabbed from the net and decoded is in such a format. I didn't lie to you. Assuming that the person who posted it is honest, yes, that is the format. > I grabbed that file, and it was HUGE. >At 8 kbytes/second of voice, I am not sure if anyone can tolerate >the traffic. > Be real. Ten years ago people were saying that bit-mapped graphics took an incredible amount of memory/storage/whatever and would never catch on. Voice takes more bits than text. Any attempt to alter this degrades voice quality and increases system complexity. In a local net of workstations with large disks, it is not worth the trouble to compress voice. >A lot of foreign e-mail addressees pay $$$ per kilobyte of >messages RECEIVED and transmitted. I think Japanese users pay >something like $1.00/kilobyte. That comes to $8/second! > Then don't send voice mail to Japan. You don't run NFS servers from Japan, do you? >Also, with straight PCM, the much touted DSP chip isn't necessary. Yup. I bet it comes straight out of the codec. >Surely, something like LPC (linear prediction coding) or PARCOR >(I forgot what that stands for) can be implemented easily with >the DSP chip. It would be simple to convert to ADPCM, and cut the data size in half. You wouldn't even need the DSP. Who knows, someone might do this in a mail bridge. Unfortunately, the format does not seem to allow the type of coding used to be specified, which seems to eliminate the possibility of compressing voice mail when sent over expensive routes. > >Izumi Ohzawa >izumi@violet.berkeley.edu >...ucbvax!violet!izumi Sorry I seem a bit testy in this posting. I'm sure that someone argued once that you'd never get an automobile to market because it weighed too much to go on a horse-drawn trailer. Or something like that. I think email charges to Japan have about as much relevance to how you would design a good voice-mail system as horse freight does to automobiles. Peter Desnoyers desnoyer@apple.com