Dear Mr. Yamamoto,
Thank you for reporting your findings on G.726. As an Associate Rapporteur of Q10/16 and Editor of G.191 (STL), I would like to give some feedbacks on this report. Please accept my apologies for this slow response.
Firstly, to give you some idea on the narrow-band codecs available in G.7xx series, those codecs usually expects band-pass filtered signal between 300 and 3400 Hz, and are characterized using speech signal (and perhaps modem signal) that lies within that frequency range.
This means that extremely low frequency input signal such as 1 and 2 Hz is not expected. For the case of G.726, there is a backward adaptive predictor. I do not have your signal and might be wrong in saying this, but my assumption is that feeding a low-frquency signal puts the codec state in an extreme divergence that results in large amplitude impulses as you report.
My suggestion for this would be to have either a band- or high-pass filter that eliminates signal below 300 Hz before the encoding block. An example implementation of such filter is also available in STL.
I would be happy to answer any further questions, but I think it is appropriate to discuss it at WP3 audio reflector (wp3audio@yahoogroups.com).
Best regards, yusuke -- ============ Yusuke Hiwasaki, NTT Cyber Space Labs. hiwasaki.yusuke@lab.ntt.co.jp Tel: +81-422-59-4815 (Fax: +81-422-60-7811)
BEGIN --- 8< --- CUT HERE --- 8< --- From: itu-sg16-bounces@lists.packetizer.com [mailto:itu-sg16-bounces@ lists.packetizer.com] On Behalf Of Yamamoto, Tetsuya Sent: Friday, 16 July 2010 8:46 To: itu-sg16@lists.packetizer.com Subject: [itu-sg16] Question on G.726
Dear Q10/16 Experts,
Sorry to bother you.
We have found unexpected results when using G.726.
When particular waveform audio data are input to G.726 encoder, the decoded data from G.726 decoder
include large amplitude impulses.
We can reproduce the same phenomena at any of 16, 24, 32 or 40kbps bitrate when we put sine wave signal
with very low frequency such as 1Hz or 2Hz into G.726 codec.
Our codec module is generated from the C source code provided by G.191 software tools as is
and the all test sequences provided by G.726 Appendix II are passed with this module.
It would be appreciated if anyone could give me some suggestions regarding this matter.
Thanks and best regards,
Tetsuya Yamamoto
Sony Corporation END --- 8< --- CUT HERE --- 8< ---