Why this is fast
Native server ffmpeg runs roughly 4–8x faster than browser WebAssembly on the same job because it uses real SIMD, threading, and disk I/O. The video stream is discarded entirely (-vn), so even a 4K MP4 only costs the audio encode. Uploads land at the nearest Backblaze B2 point of presence via a presigned URL (regional latency, not US-east tax), the MP3 comes back through Bunny CDN from the same edge. Most extractions finish in well under a second of CPU time.
What this tool does, exactly
We run ffmpeg -i input -vn -c:a libmp3lame -q:a 2 -f mp3 output.mp3. The -vn tells ffmpeg to ignore every video stream in the source, so the encode only touches audio. LAME runs at VBR quality 2 (~190 kbps average), which is perceptually transparent for speech, music, and screen-recording audio. The accepted container list is video/mp4, video/quicktime, video/webm, and video/x-matroska — if the file has audio, this works.
MP3 today, WAV coming — the honest status
The V1 server spec pins one output container per slug, and video-to-audio currently ships MP3. WAV (uncompressed PCM, the right choice when the audio is headed into an editor or DAW) lands when the spec layer supports per-options containers — the work is queued. Until then, if you need lossless WAV for Audacity, Logic, Premiere, or Resolve, a desktop ffmpeg does it in one command. For sharing, podcasting, transcription, or chat — all the cases where MP3 is the right answer — this page is ready.
Sister tools
Specifically converting an MP4? MP4 to MP3 converter is the focused version. Converting between audio formats already on disk? MP3 converter and audio converter both accept WAV, M4A, FLAC, OGG, and more. Trim the extracted MP3? MP3 cutter. Shrink it for email or Discord? Compress audio. Clean up background hiss in the extracted audio? Remove background noise. If you only need a transcript, record on Clipy instead — every recording auto-transcribes and you can skip the audio-extraction step entirely.