Re: Bridging the digital voice and data gap

"John D. Hays" <john@...>

Hi Tyrell,

Vocoders and other codecs are different beasts.  

Two short articles:

In layman's terms, a codec converts an analog stream into a coding format.  Some coding formats are lossless and produce a true representation of the orignal analog stream in a digital format, but most such as MP4 and AAC, use compression to get a lower data rate.

A vocoder is considered a type of codec, in that it is converting "analog" voice to what is usually a very low data rate bitstream and back.  However, it's methodology is fundamentally different than traditional audio lossy and lossless codecs.  AMBE and CODEC2 both approach the problem by detecting and recognizing phonemes (the basic sounds that make up human speech) and such things as frequency slope, amplitude, and inflection in the analog stream.  The vocoder then represents the phonemes and modifiers as digital codes (e.g. 0x3E might represent the 'th' sound), which are transmitted digitally, where the receiving end uses those codes to synthesize the original speech.  This process is focused on human speech and has the side effect of mostly ignoring sounds that are not human speech.

The semantic is important to differentiate both the technique and result of the method of encoding and decoding speech.  If you transcoded an MP3 (lossy) music stream into AMBE or Codec2 the result would likely be unrecognizable.  

It may be that transcoding between two similar vocoders would exhibit less deterioration of a voice signal, since they would both be looking for the same attributes in the "audio" signal.  We won't really know until someone tries it in the real world.

Everything else on the "data" side of a complex DV radio protocol can probably be handled, in one form or another, though there would be mismatches.  For example, going from D-STAR to DMR, "callsign addresses" and unit identifiers could be mapped to one another.  On the other hand, there is no equivalent to D-STAR's ancillary data (e.g. comment text, D-PRS, etc.) in IRLP's network, though a SIP link could send it as a message in the control channel.

John D. Hays
PO Box 1223, Edmonds, WA 98020-1223 

On Fri, Aug 3, 2012 at 7:00 AM, Tyrell Berry <kd7kuj@...> wrote:

Again, I'm not an audio engineer, and I have been wrong before...  but isn't saying that the chip is a vocoder and it is not compression kind of like saying MP3 is a Muscoder and is also not compression? Is it somehow true that because someone chooses to use a different vocabulary that the principles of which I speak somehow cease to apply?

It has been said several times before, and I will agree: it's entirely possible that transcoding through back to back vocoders may produce legible audio...  Only one way to find out, and I look forward to hearing recordings of the results.  

On Aug 3, 2012 6:51 AM, "Bryan Hoyer" <bhhoyer@...> wrote:

The UDR56K has an expansion interface which is designed for an optional ambe vocoder (it is NOTcompression).

DVSI chips are TI DSPs that are factory programmed with their code. The interface is 2 SSPs or synchronous serial port which connect directly to the Marvell SOC.

The interface and drivers wil be documented.


Join to automatically receive all group messages.