similar to: Fwd: [codec] Last proposal for the update draft

Hi Marcello, It turns out that the problem has a much simpler explanation. As far as I can tell, it's actually a bug in silk_sum_sqr_shift() and this trivial patch appears to fix the problem: http://jmvalin.ca/misc_stuff/sum_sqr_shift_fix.patch It would still require some testing to check that the fix doesn't have any bad side effect. Let me know how well the fix works for you. Again,

Hi Marcello, Actually, we were careful to avoid the undefined behaviour here. In fact, we are specifically running a clang test detecting undefined behaviour. If you look at the silk_SMLABB_ovflw() macro, you will see it is based on silk_ADD32_ovflw(), which is defined as: #define silk_ADD32_ovflw(a, b) ((opus_int32)((opus_uint32)(a) + (opus_uint32)(b))) By casting to unsigned, all the cases

Right, there shouldn't be a problem with undefined behavior. That said, a 64 bit implementation will work very well - in fact that's how it was done originally. The reason for the current implementation is to minimize 64-bit operations in order to improve performance on limited-width architectures. This functions gets used extensively, and I think the current implementation is faster on

Hi Jean-Marc, well spotted! The patch provided fixes the issue for me. Nevertheless, in my code (and I would suggest doing the same in libopus) I am going to replace the function with one that accumulates on 64 bits and then calculates the shift, for at least 4 reasons: - It is less and simpler code - The result is likely to be slightly more accurate in case big numbers come early in the

Hi Drew, On 11/13/2017 03:34 PM, Drew Allen wrote: > Does your encoder support ch253/254? Not at this point. Only families 0 and 1 are supported. I think adding family 2 shouldn't be relatively straightforward, but adding family 3 would require some thought. Cheers, Jean-Marc > Cheers, > Drew > On Mon, Nov 13, 2017 at 12:33 PM Jean-Marc Valin <jmvalin at jmvalin.ca >

Hi Andreas, So I encoded your file in chunks with a slightly modified version of opusenc_example and I can't hear anything wrong. Maybe there's a problem in the tools you used? I uploaded the files at: https://jmvalin.ca/misc_stuff/continuous.opus (one file) https://jmvalin.ca/misc_stuff/continuous.wav (one file, decoded) https://jmvalin.ca/misc_stuff/chained.opus (many small files)

Hi, Just to let everyone know that I released libopusenc version 0.1.1. It does not change the API/ABI compared to 0.1, but fixes a few bad bugs. You can get it from https://www.opus-codec.org/downloads/ For those interested in testing it, there's an experimental patch to make opus-tools use libopusenc for the encoder:

Hi Jean-Marc, thank you for looking into this. I'm using the current release version of libopus (1.2.1) and version 0.1.10 of opus-tools, both built from source. Libopusenc is the current git master. Indeed I cannot hear any glitches in continuous.opus, continuous.wav, and chained.wav. chained.opus decodes just fine with opusdec as well, but doesn't play properly in any standard audio

Hey Jean-Marc, Does your encoder support ch253/254? Cheers, Drew On Mon, Nov 13, 2017 at 12:33 PM Jean-Marc Valin <jmvalin at jmvalin.ca> wrote: > Hi, > > Just to let everyone know that I released libopusenc version 0.1.1. It > does not change the API/ABI compared to 0.1, but fixes a few bad bugs. > You can get it from https://www.opus-codec.org/downloads/ > > For

Hi Jean Marc, please find attached the audio file (mono 16khz). I shortened it to about 10 seconds. I also add 2 patches that worked for me. Further info that might help: - The problem seems to be related to silk_burg_modified not reaching the maximum gain, so the actual filter order is 16 rather than 2 (which is what would be expected with a sine wave). - The problem seems to happen when

Hi, Before I comment on the graphics you posted to visualize the difference between two audio signals, I'd like to ask for your help in evaluating my JPEG encoder. I've encoded an image with JPEG and then computed the difference with the original. I then converted the difference to sound. You can listen to the image difference on this clip: https://jmvalin.ca/misc_stuff/diff.wav Can you

Currently we use non-cryptographically secure random numbers in two places, the error filter (to inject errors at random) and the random plugin. For this we have used either random_r or a home-brew-ish Linear Congruential Generator. Use of random_r is problematic on BSDs because it doesn't exist there. Use of the LCG is simply a bad choice. Replace both uses with a better quality and

Currently we use non-cryptographically secure random numbers in two places, the error filter (to inject errors at random) and the random plugin. For this we have used either random_r or a home-brew-ish Linear Congruential Generator. Use of random_r is problematic on BSDs because it doesn't exist there. Use of the LCG is simply a bad choice. Replace both uses with a better quality and

Just to be clear, my goal here wasn't to make fun of anyone, but to drive the point that spectrograms should *never* be used to demonstrate quality. The only case where they can sometimes be useful is for diagnostic purposes. If you hear something and you're not sure what you're hearing exactly, then sometimes a spectrogram can help you figure out what it is. That's pretty much it.

Jean-Mark sarkasm. Jean-Markasm. (Bonus points for providing an actual noisy WAV! ^_^) On 30/10/2017 20:28, Jean-Marc Valin wrote: Hi, Before I comment on the graphics you posted to visualize the difference between two audio signals, I'd like to ask for your help in evaluating my JPEG encoder. I've encoded an image with JPEG and then computed the difference with the original. I then

Hi Mark, Jean-Marc, Thanks for your comments. On Sun, Jun 12, 2016 at 6:34 AM Mark Harris <mark.hsj at gmail.com> wrote: > Hi Felicia, > > A few comments: > > > - /* CELT can only support up to 20 ms */ > > subframe_size = st->Fs/50; > > - nb_subframes = frame_size > st->Fs/25 ? 3 : 2; > > + nb_subframes =

Hi guys, as MP3 and Opus have very similar objectives, I think the original poster's question was a valid one: Why does Opus have more artefacts in the lower frequency ranges than MP3 has? The spontaneous suspect that lower frequency artefacts may be more noticeably than higher frequency artefacts seems plausible, also. Is it a matter of energy (which is higher for higher frequencies)? When

Hi! A note on style: Looking at this chunk of the patch -- @@ -382,9 +382,15 @@ int main(int argc, char *argv[]) frame_size = sampling_rate/25; else if (strcmp(argv[ args + 1 ], "60")==0) frame_size = 3*sampling_rate/50; + else if (strcmp(argv[ args + 1 ], "80")==0) + frame_size = 4*sampling_rate/50; +

Hi everyone, I'm pleased to announce version 0.6.0 of CELT, with many quality improvements. This includes better stereo coupling, better handling of transients, and better handling of highly tonal signals. In addition, packet loss robustness has been improved through the optional use of independent (intra) frames. CELT now supports a larger dynamic range, suitable for encoding 24-bit audio

Thank you. Very helpful. > On Jan 27, 2017, at 12:40 PM, Jean-Marc Valin <jmvalin at jmvalin.ca> wrote: > > On 27/01/17 12:29 PM, Jon Lederman wrote: >> Thank you. Yes, we do need both channels independent. So, if we >> encode each channel separately, we will be sacrificing the >> compression ratio we would achieve with stereo encoding, correct? > > Not