Hi > Actually the effectiveness of the 40ms frame rate is something of a > surprise, in Speech Coding 101 I was taught that it shouldn't work. It > would be good to investigate this further.
Hmm, curious. I haven't yet listened to live 1,200, but it clearly implies that the error levels are similarly accurate when carrying forward previous predictions as they are to use fresh predictions. Whilst on the surface that seems a little sad, I guess the point is to use that knowledge. Given that amplitude is such a large part of the bit budget, what about partial amplitude updates on each frame, eg not all predictions updated on every 20ms frame, eg predict some changes, transmit others. ie you can get 20ms updates on the most important changes and 40ms updates on the rest? > Predictive LSP coding is > something Jean-Marc implemented a few months back and code exists for > that. Still some testing and optimisation reqd. Are there any recorded samples? Does it improve the perceived quality? Sorry, probably I read this on the list already, is this what he calls "codebook" changes? > On my Codec 2 page http://www.rowetel.com/blog/?page_id=452, 'Progress > to Date' section dot point (6) there are samples of what Codec 2 sounds > like with original phases. Consider that the upper quality limit. Hmm, interesting. Are you the "male" voice? It seems to work extremely well for your voice. My results using asterisk (talking to male (me), female (wife) and two young kids (one of each)) are pretty decent, but regularly drift towards the samples you label as "here is a counter example". It's rarely as bad as that specific example, but I guess it's that zing/buzz which is very audible on that sample, which I hear coming in and out on the odd syllables when I talk. So for me it's more like fairly good for 5 syllables, then zings a bit for the next syllables, then back to good. Clearly I need to take some time to do some good recordings of this or it's just heresay! One good, but bad feature of the asterisk setup is that I can record all the conversations, but only the codec2 side... Means I don't have the original audio. I guess at this point it would be useful to get really constructive about voice samples. What format and how best can "us users" submit interesting snippets? Should we choose a standardised and slightly longer paragraph that different people could record to illustrate quality variation? ("round the ragged rock the ragged rascal ran.."). The question is more about how to deliver something which can be usefully analysed and isn't just a curiosity? Any suggestions from the top? All in all it's very exciting progress. Reducing the zing/buzz ever so slightly at 2.4 would for me bring it up a big level in perceptable quality. And having a 4kbit ish toll quality codec would be very interesting for my day to day usage over voip (obviously need to increase the frame size for best effect) Thanks for all your work! Ed W ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
