duration is wrong for m4a audios but works for flac and mp3 files

🚀 The feature

Currently torchaudio (try torchaudio.info(m4a_path), regardless of version returning wrong results.
I wish we could get duration correctly for m4a files.

Motivation, pitch

I want to do segmentation of m4a audios but it can't evaluate correctly the duration of audios.

I was using this script https://github.com/k2-fsa/text_search/blob/master/examples/libriheavy/run.sh
and the first stage is prepare_manifest.py https://github.com/k2-fsa/text_search/blob/master/examples/libriheavy/prepare_manifest.py
it works well when audio files are ".flac" but when I am using dataset that is m4a audio format it is computing duration wrong.

Alternatives

1. https://github.com/lhotse-speech/lhotse/issues/1121
   pzelasko was able to create a work around it for now.

2. I can compress m4a audios to opus type and then do segmentation i guess.

Additional context

No response

Start by reproducing the incorrect result with torchaudio.info(m4a_path) and compare it with the working FLAC and MP3 cases. Read examples/libriheavy/prepare_manifest.py and run.sh to see how duration is consumed, then review the workaround referenced in lhotse issue 1121. Done means m4a duration is computed correctly for segmentation.