[FFmpeg-user] How can I get byte-accurate segments?

Moritz Barsnick barsnick at gmx.net
Fri Feb 12 12:23:52 CET 2016

On Fri, Feb 12, 2016 at 11:02:32 +0100, Paul B Mahol wrote:
> ffmpeg -f lavfi -i
> anoisesrc=sample_rate=16000:duration=35:nb_samples=16000 -map 0:a -c:a
> pcm_s16le -ac 1 -f segment -segment_time 5 -segment_format s16le
> out.%03d.raw

Well, anoisesrc was just an for creating arbitrary input. The original
poster may not be able to influence the sample rate and number of
samples in his input. Therefore I sought after other flags to influence

> You can't resample after, you will get different number of samples
> per packet.

So, in the real world: Would you recommend inserting the aresample
filter? (And possibly something for downmixing to mono.)

My head is spinning. Samples, packets, frames. How do they relate in
audio? Especially regarding this codec.

I would have thought that, in the path from codec to muxer, the concept
of frames would be somewhat broken down, and (for this particular
codec) there would be a frame per sample. Or is it a packet per sample?
Or neither of the two?

Anyway, I would have thought that this codec, being "key frame only",
could and would be cut at exact points, assuming the sample rate allows
an exact number of samples to fit into the given time interval.

Carl Eugen's response was better than both of mine: Just try to solve
the problem, don't try to understand all the mechanisms. ;-)


More information about the ffmpeg-user mailing list