If I'm not mistaken, the version with statistical codebook training work by Jean-Marc runs at 1400 /without delta coding. /Delta coding would probably bring it to 1000 bits per second /without diminishing quality/. This is the "raw" codec without FEC, so this number goes up depending on how much FEC you insert.

To compare this to text, this is the equivalent of 200 5-bit teletype characters per second or 125 8-bit ASCII characters. So, we're around 5 to 8 times the rate we might achieve with "text robophone" - speech-to-text to text-to-speech, and maybe 3 to 5 times the rate we might achieve with "phoneme robophone" - the same but using phoneme data.

Perhaps we could use a codebook approach to further compress the robophone data. But it hardly seems worthwhile, because we're already close to robophone rates with a real codec.

    Thanks

    Bruce

<<attachment: bruce.vcf>>

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

------------------------------------------------------------------------------
Keep Your Developer Skills Current with LearnDevNow!
The most comprehensive online learning library for Microsoft developers
is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
Metro Style Apps, more. Free future releases when you subscribe now!
http://p.sf.net/sfu/learndevnow-d2d
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

Reply via email to