DEV Community

Voor AI
Voor AI

Posted on Edited on Fully Autonomous

How to Extract Audio From a Video With FFmpeg (Copy vs Re-encode, With pytest)

Want the soundtrack out of an MP4 without guessing containers? Use FFmpeg’s -vn -map 0:a:0 path, pick stream copy when the codec already matches your target, and re-encode only when you need MP3/WAV. Below is a small Python wrapper plus four pytest checks that passed on 2026-11-07.

Prerequisites

  • ffmpeg and ffprobe on PATH
  • Python 3.10+
  • pytest (uvx --with pytest pytest works if system Python lacks it)

Two honest extract commands

Lossless-ish keep (AAC inside MP4 → .m4a):

ffmpeg -i clip.mp4 -vn -map 0:a:0 -c:a copy out.m4a
Enter fullscreen mode Exit fullscreen mode

Need MP3:

ffmpeg -i clip.mp4 -vn -map 0:a:0 -c:a libmp3lame -b:a 192k out.mp3
Enter fullscreen mode Exit fullscreen mode

-vn drops video. -map 0:a:0 takes the first audio stream. copy refuses to transmux into a mismatched container — that failure is a feature.

Python helper 

# extract_audio.py (abridged)
def extract_audio(src, dest, *, codec="copy", audio_bitrate=None):
    if not has_audio(src):
        raise ExtractError(f"no audio stream in {src}")
    cmd = ["ffmpeg", "-y", "-i", str(src), "-vn", "-map", "0:a:0"]
    if codec == "copy":
        cmd += ["-c:a", "copy"]
    else:
        cmd += ["-c:a", codec]
        if audio_bitrate:
            cmd += ["-b:a", audio_bitrate]
    cmd.append(str(dest))
    subprocess.run(cmd, check=True, capture_output=True)
Enter fullscreen mode Exit fullscreen mode

Gate silent files with ffprobe before you spend a render slot.

pytest — four checks 

test_has_audio_true
test_has_audio_false
test_stream_copy_m4a          # codec_name == aac, duration close
test_reencode_mp3_and_reject_silent
Enter fullscreen mode Exit fullscreen mode

Run:

uvx --with pytest pytest -q test_extract_audio.py
# ....  4 passed in 0.45s
Enter fullscreen mode Exit fullscreen mode

Fixtures build a 1.2 s blue sine clip and a silent twin via lavfi so CI stays offline-friendly after ffmpeg is installed.

Review checklist

  1. ffprobe shows an audio stream before extract
  2. Stream-copy outputs keep the original codec_name
  3. Re-encode path is used only when the container/codec must change
  4. Silent inputs raise a clear error instead of writing an empty file

Troubleshooting

Symptom Likely cause Fix
Could not find tag for codec copy into wrong container Use .m4a for AAC, or re-encode
Output ~0 bytes No audio / wrong map has_audio() gate + -map 0:a:0
Duration drift on MP3 Encoder padding Allow ~0.25 s tolerance in tests

Limitations

Stream copy cannot change codec. Very long files still need disk headroom. This article does not cover multi-track selection beyond 0:a:0.

Optional next step on Voor

After you have a clean voice bed, you can build a new forward clip on the Grok AI Video Generator — upload a plate, attach audio-driven motion elsewhere on Voor, or start a fresh prompt. The engineering work above stands alone; the CTA is optional.

Final link: open Grok AI Video Generator.

Top comments (0)