[FFmpeg-user] Audio transcoding with hardware acceleration

[Nikita Zlobin]

In Thu, 2 Jun 2022 18:51:23 +0000 (UTC)
james young via ffmpeg-user <ffmpeg-user at ffmpeg.org> wrote:

> Hi,I am working on a software which converts the audio to mp3 with 8k
> bitrate using ffmpeg. The audio can be of any codec and the output
> will be mp3. This conversion takes longer time to finish for codecs
> like opus. So I am intending to use the gpu based machines to speed
> it up. So far i have tried this guide
> https://docs.nvidia.com/video-technologies/video-codec-sdk/ffmpeg-with-nvidia-gpu/
> which tells to build ffmpeg with --enable-cuda-nvcc --enable-libnpp
> options. Performing the configuration in this guide will only allow
> the video conversion to use the hardware acceleration. Hardware
> acceleration is not being used for audio conversion. Is hardware
> acceleration supported for audio conversion? Regards,James
> _______________________________________________ ffmpeg-user mailing
> list ffmpeg-user at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-user
> 
> To unsubscribe, visit link above, or email
> ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".

Besides nv, there's also family of video opencl plugins, with
"program-opencl" among them. If you are using API, you could make
verbatic A->V conversion before chain (just by copying
audio data into 1-row video) and vice versa. Some depth bits could be
moved into size domain (I'm not sure though, which dimention is best to
encode depth in opencl, since I never worked with opencl). Could be
even so, that opencl doesn't restrict texture size at all, unlike
vaapi, vdpau, opengl.

Without programmatic approach - the only solution I see is using
predefined A->V and V->A conversion plugins, like waveshape generator
for input and... well, I see only spectrumsynth for output. Although
such power waste could at least compensate opencl performance boost (if
not make it worse).