Benjamin Larsson <[email protected]> added the comment:
Mark wrote:
> New submission from Mark <[email protected]>:
>
> With some flv files using the nellymoser codec, ffmpeg does not
> correctly identify the sampling frequency. ffmpeg then produces files
> with low and deep audio. Here are a couple examples of files ffmpeg
> won't correctly convert:
>
> http://video.whothaman.net/010/250/10250/10250.orig
> http://video.whothaman.net/010/969/10969/10969.orig
>
> Using a tool called yamdi to insert metadata fixes the problem. It's at
> http://yamdi.sourceforge.net/
>
This issue could be valid.
The demuxer must let the codec probe sample rates using this code if the
stream is identified to have this sample rate:
Stream #0.1: Audio: nellymoser, 5512 Hz, mono,
168 <#l168> switch (buf_size) {
169 <#l169> case 64: // 8000Hz
170 <#l170> blocks = 1; break;
171 <#l171> case 128: // 11025Hz
172 <#l172> blocks = 2; break;
173 <#l173> case 192: // 16000Hz
174 <#l174> blocks = 3; break;
175 <#l175> case 256: // 22050Hz
176 <#l176> blocks = 4; break;
177 <#l177> case 512: // 44100Hz
178 <#l178> blocks = 8; break;
179 <#l179> default:
180 <#l180>
av_log(avctx, AV_LOG_DEBUG, "Tag size %d.\n", buf_size);
181 <#l181> return buf_size;
182 <#l182> }
My proposed fix to this issue is to let the demuxer set the samplerate
to 0 when it finds a 5512 samplerate for nellymoser. And then add a
check in nellymoserdec that checks if the samplerate is 0 and if it is
triggered set the samplerate based on the blocksize based on the code
from above.
MvH
Benjamin Larsson
_____________________________________________________
FFmpeg issue tracker <[email protected]>
<https://roundup.ffmpeg.org/roundup/ffmpeg/issue1319>
_____________________________________________________
