Extracting audio from a video
Pulling the soundtrack out of an MP4 or MKV without degrading the signal: which format to pick, which bitrate, and the traps worth knowing about.
An hour-long interview filmed on a phone takes up 3 to 5 GB on the memory card. The soundtrack itself weighs around fifty megabytes. If all you need is the audio — for a podcast, a transcript, an edit, or simply to listen to in the car — carrying the picture around serves no purpose.
A video file is a container, not a format
MP4, MOV, MKV, AVI, WebM: these are containers. Inside, a video track (usually H.264 or HEVC) sits alongside one or more audio tracks — almost always AAC-LC, sometimes AC-3 on files that came off a digital TV recording or a DVD, uncompressed PCM on the rushes from certain cameras — and sometimes subtitles.
Extracting the audio means opening that container and taking the sound track out of it. Two very different operations hide behind the same word:
- Stream copy: the track is copied across untouched into a .m4a or .aac file. No loss, and the job is near-instant since there is nothing to recalculate.
- Re-encoding: the track is decoded then recompressed, typically to MP3. The result plays absolutely everywhere, but you are stacking a second round of compression on a signal that has already been through one.
That distinction explains most of the disappointment. An MP3 at 128 kbps made from an AAC track at 128 kbps sounds worse than the original even though the stated bitrate is identical: the two codecs throw away different information, and their artefacts add up. The sound is only genuinely intact if the tool copies the track without touching it, so look for a stream-copy option, or an M4A/AAC output that states explicitly that nothing is re-encoded. If the tool re-encodes — which is the usual case — you are in for another generation of compression whether the output is called AAC or MP3. When that happens, choose on the basis of the compatibility you need and compensate with a more generous bitrate.
MP3, AAC or WAV: what the choice actually changes
| Output | Size for 60 minutes | When to choose it |
|---|---|---|
| MP3 128 kbps | ≈ 58 MB | Speech, sending by email, universal playback |
| MP3 192 to 256 kbps | ≈ 86 to 115 MB | Music, or when MP3 is the only option |
| AAC / M4A 128 kbps | ≈ 58 MB | Better quality than MP3 at the same bitrate, Apple ecosystem |
| WAV 48 kHz, 16-bit, stereo | ≈ 690 MB | Editing, archiving, later processing |
One detail almost nobody checks: the audio track in a video is most often sampled at 48 kHz rather than the 44.1 kHz of a CD: 48 kHz has been the broadcast standard since the DV era. Not every file respects it — some phones and screen-recording tools capture at 44.1 kHz. Check the sample rate of your source, then keep it on the way out: any deviation forces a resample, one more operation with no benefit whatsoever here. MP3 handles 48 kHz perfectly well, so if your tool forces 44.1 kHz on the output it is resampling for nothing. Keep the original rate.
Getting the track out without installing anything
The Extract Audio tool accepts MP4, MOV, AVI and MKV files and offers MP3, WAV or AAC output. If you already know it is going to be MP3 and nothing else, MP4 to MP3 goes straight to the point.
- Drop the video into the upload area.
- Choose the output format and bitrate using the table above.
- Start the conversion: allow a few seconds for every ten minutes of video, with most of that time going on the upload rather than the processing.
- Download the audio file.
Both tools run on the server: demultiplexing several hundred megabytes inside a browser tab would hammer the machine for nothing. The file is uploaded, processed, then deleted from our servers after one hour. Every visitor gets one free conversion a day, with no sign-up.
If your video is over the maximum size shown on the tool page, there are two ways out: bring its size down first with Compress Video — bearing in mind that the audio track will be re-encoded in the process too, so that is one extra generation of loss — or split the video in two in an editing package and handle each half separately.
The traps that spoil the result
More than one audio track in the same file
An MKV of a film will often carry the original soundtrack and a dubbed version. A multi-camera shoot may have the lapel mic on one track and the camera mic on another. An online tool takes the first track declared in the container, which is not necessarily the one you want. Open the file in a player that can list the tracks before you start the extraction, and work out which one you are after.
The sound only comes out of one ear
A classic with a lapel mic plugged into the left input: the track really is stereo, but the right channel is empty. Extraction corrects nothing: the fault is already in the source. The cure is a switch to mono. In WAV the size halves automatically; in MP3 or AAC it will not budge unless you drop the bitrate as well, which mono lets you do with no audible loss on speech. One warning: mixing a track with one empty channel down to mono loses around 6 dB of level, to be made up in the edit.
A video that simply has no sound
Screen recordings made without system-audio capture switched on, silent exports from certain editing tools and GIFs converted to MP4 contain no audio track at all. Extraction fails because there is nothing to process. A related but distinct case: a track that is present but silent — mic muted, gain at zero — will give you a file of normal size and perfect silence. Neither is a fault in the tool: check first that the video makes a sound at your end, headphones on.
The sound-to-picture offset
Some MP4s declare a start delay on the audio track through an edit list. Once the track is isolated that delay disappears, and the sound starts earlier than it did in the original video. Harmless if you are listening to the file on its own, awkward if you plan to lay it back against the picture: note the offset before you separate the two.
A file with no labels
An extracted track arrives with no ID3 tags: no title, no artist, no artwork. On a mobile player it lands under “Unknown” and becomes unfindable three days later. Fill in the metadata afterwards if the file is going to join a music library.
One last point, and not a technical one: extracting the soundtrack from a copyrighted work in order to republish it is a copyright matter, whatever tool you used. For your own rushes, a talk you hold the rights to, or strictly private use, the question does not arise.
The right settings for what you are doing with it
- Podcast, interview, recorded lecture: mono MP3, 96 to 128 kbps, 48 kHz. Speech gains nothing beyond that, and an hour fits into 43 to 58 MB.
- Music, live recording, sound design: AAC copied as-is if the tool allows it, otherwise MP3 at 256 kbps. Below 192 kbps, cymbals and synth pads give the compression away.
- Editing or post-production: WAV at 48 kHz / 16-bit. Heavy, but it is the only one of the three outputs that does not add a layer of compression on every round trip between applications (a compressed lossless format such as FLAC would do the same job at roughly half the size, wherever your editing workflow accepts it).
- Automatic transcription: mono is more than enough, since most recognition engines bring the signal down to 16 kHz before analysis anyway. Handing them a 700 MB WAV changes nothing but the upload time.
- Ringtone or short clip: AAC/M4A, clearly better than MP3 at low bitrates.
When installed software is still the better option
Three situations justify leaving the browser: batch-processing dozens of files, keeping all the audio tracks separately, and files big enough that the upload takes longer than the extraction itself. A command-line utility does this very well, at the cost of an installation and a syntax to learn.
For a one-off extraction, or from a machine where you cannot install anything — work laptop, shared computer, tablet — the online version is still the shortest route. And if your problem turns out not to be the soundtrack but the size or the compatibility of the video file itself, look instead at converting the video format.
If extraction becomes a habit rather than an occasional fix, the Convertu subscription at €7 a year lifts the one-conversion-a-day limit across all the server-side tools.