Hi, Sorry to be posting about a subject that may have already been answered. If so, please point me in the right direction. I'm developing a dictation application on the Dell Axim (Windows Mobile 5.0 Pocket PC). A key requirement of the application is the best possible sampling rate as the audio goes into a speech reco system. So, I've set up my wrapper around libspeex to capture audio (1 channel, 22050Hz, 16 bits/sample) in blocks of the frame size. However, the encoder lags significantly - typically upwards of 2000 frames, at complexity <= 1 and quality <= 3. I've a On the decode side, the decoder is not able to do real time at these sampling rates (16kHz, 22kHz and 32kHz). The only sampling rate that seems to work at real time is 8kHz for decode. However, encode at 8kHz is still nowhere near real time. Is there anything I can do to get to the higher sampling rates and real time? Or is this just not possible? FWIW, the basic app uses MFC and the encode and decode are done in separate threads. Thanks much, Sunil --------------------------------- Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates starting at 1?/min. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/speex-dev/attachments/20060819/23174356/attachment.html
Hi there, i'm not into speex internals but i feel that your problems are due to the fact that speex doesn't do yet fixed point operations and it seems that you lack FPU on your PDA. So i would expect a big boost in performances when the fixed point version of Speex will be out. Cheers, Luca On 8/20/06, Sunil <dualmax88@yahoo.com> wrote:> > Hi, > > Sorry to be posting about a subject that may have already been answered. > If so, please point me in the right direction. > > I'm developing a dictation application on the Dell Axim (Windows Mobile > 5.0 Pocket PC). A key requirement of the application is the best possible > sampling rate as the audio goes into a speech reco system. So, I've set up > my wrapper around libspeex to capture audio (1 channel, 22050Hz, 16 > bits/sample) in blocks of the frame size. However, the encoder lags > significantly - typically upwards of 2000 frames, at complexity <= 1 and > quality <= 3. I've a > > On the decode side, the decoder is not able to do real time at these > sampling rates (16kHz, 22kHz and 32kHz). The only sampling rate that seems > to work at real time is 8kHz for decode. However, encode at 8kHz is still > nowhere near real time. > > Is there anything I can do to get to the higher sampling rates and real > time? Or is this just not possible? > > FWIW, the basic app uses MFC and the encode and decode are done in > separate threads. >-------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/speex-dev/attachments/20060821/d2ef50f6/attachment.html
Thanks Luca. Turns out I was not as thorough in my search as I thought I was. The latest sources (at least from the svn repository) do in fact have a FIXED_POINT mode for ARM processors on Windows Mobile 5.0. Using this turned out to be a godsend. I can now encode and decode 22kHz mono at complexity and quality <= 3 in pretty much real time (there seems to be a 2 second lag in encode, but that's not troublesome), which suits me just fine. Thanks again! Luca De Cicco <ldecicco@gmail.com> wrote: Hi there, i'm not into speex internals but i feel that your problems are due to the fact that speex doesn't do yet fixed point operations and it seems that you lack FPU on your PDA. So i would expect a big boost in performances when the fixed point version of Speex will be out. Cheers, Luca On 8/20/06, Sunil <dualmax88@yahoo.com> wrote: Hi, Sorry to be posting about a subject that may have already been answered. If so, please point me in the right direction. I'm developing a dictation application on the Dell Axim (Windows Mobile 5.0 Pocket PC). A key requirement of the application is the best possible sampling rate as the audio goes into a speech reco system. So, I've set up my wrapper around libspeex to capture audio (1 channel, 22050Hz, 16 bits/sample) in blocks of the frame size. However, the encoder lags significantly - typically upwards of 2000 frames, at complexity <= 1 and quality <= 3. I've a On the decode side, the decoder is not able to do real time at these sampling rates (16kHz, 22kHz and 32kHz). The only sampling rate that seems to work at real time is 8kHz for decode. However, encode at 8kHz is still nowhere near real time. Is there anything I can do to get to the higher sampling rates and real time? Or is this just not possible? FWIW, the basic app uses MFC and the encode and decode are done in separate threads. __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/speex-dev/attachments/20060821/20740c2a/attachment.htm
> I'm developing a dictation application on the Dell Axim (Windows > Mobile 5.0 Pocket PC). A key requirement of the application is the > best possible sampling rate as the audio goes into a speech reco > system. So, I've set up my wrapper around libspeex to capture audio > (1 channel, 22050Hz, 16 bits/sample) in blocks of the frame size. > However, the encoder lags significantly - typically upwards of 2000 > frames, at complexity <= 1 and quality <= 3. I've aJust a detail here. 22.05 kHz is not recommended for Speex. I suggest you use 16 kHz. Jean-Marc