Free Audio Extractor

Worked example

A 10-minute phone video with a stereo AAC soundtrack at 128 kbps and 44.1 kHz. In the default "Convert" mode with MP3 and "Match source", the bitrate button reads "Match source (192 kbps)" (128 kbps plus 25 percent for an MP3 encode, rounded up to the next step) and "Estimated output:" shows about 13.7 MB. The "Voice / speech" preset (MP3, mono, 64 kbps) gives about 4.6 MB for the same 10 minutes, and WAV would be about 100.9 MB. Switch to "Original quality" and the AAC track is lifted out unchanged as an .m4a file. No estimate is shown in that mode, because the size is whatever the track already takes up in the video.

Which settings to use, and the limits

• Choose between "Original quality" and "Convert". "Original quality" (badge "Instant") copies the audio track out untouched, so nothing is lost and there is little to wait for. The file type follows the source: AAC and ALAC become .m4a, MP3 stays .mp3, FLAC becomes .flac, Vorbis becomes .ogg, Opus becomes .opus, AC-3 becomes .ac3 and uncompressed PCM becomes .wav. "Convert" is the default and re-encodes to MP3, M4A (AAC), AAC (.aac), Opus (OGG), OGG (Vorbis), WAV or FLAC. If a track cannot be copied, the tool converts it instead and tells you.
• Pick a bitrate that fits the job. The lossy formats offer "Match source" plus 48, 64, 96, 128, 192, 256 and 320 kbps. "Match source" follows the original track's bitrate and channel count, so a small voice clip is not inflated. A higher bitrate than the source adds size, not detail. WAV is uncompressed 16-bit audio, about 10 MB per minute for stereo at 44.1 kHz. FLAC is lossless and is estimated at about 60 percent of the WAV size.
• Use the presets and sample rate for speech. "Voice / speech" sets MP3, mono, 64 kbps, which suits transcription and messaging. "Voice note (WhatsApp, Telegram)" sets Opus in OGG, mono, 48 kbps. For speech-to-text tools set Channels to "Mono" and Sample rate to "16 kHz"; the page notes that tools such as Whisper expect mono at 16 kHz. The other sample rate choices are "Same as source" and "22.05 kHz".
• Choose the audio track when there are several. If the file has more than one audio track, an "Audio track" picker lists each one (language, title, codec, channels) and ends with "Mix all tracks", which sums every track into one file. Mixing needs a re-encode, as do "Normalize loudness", "Volume boost" (-12 to +12 dB), a channel change and a sample-rate change, so "Original quality" is switched off while any of them is on. When you add several files at once, every file uses its first audio track.
• Extract just a section. "Extract a section" has "Start" and "End" sliders that move in 0.1 second steps, and the section is at least 0.1 seconds long. It is only offered for a single file; when you add several files, each one is extracted whole and the results come as a ZIP.
• Know the size cautions. A single file is read straight from disk instead of being copied into memory, so one large video is not refused up front. You get a caution above 800 MB on a normal computer, 400 MB on a computer reporting 4 GB of memory or less, and 150 MB on a phone. A memory check may also ask whether to continue. When you add several files, any file over 2 GB (1 GB with 4 GB of memory or less, 400 MB on a phone) is refused. The first use downloads the encoder, about 30 MB, which is cached afterwards. Files are processed in your browser and are not uploaded.

Common problems and how to fix them

It says "This file has no audio track to extract." and the Extract audio button is disabled.

The tool read the file's stream list and found picture but no sound, or the browser reported no audio track. Screen recordings made with the microphone off are the usual cause. Check that the original plays with sound. If the sound was saved as a separate file, extract from that file instead.

It says "Processing cancelled."

That is what you get after pressing "Cancel". Cancel stops the encoder so that a long job cannot hang the page. Press "Extract audio" again and the encoder reloads, from the browser's cache once it has been downloaded.

It says "This file couldn't be read — it may be incomplete (still downloading or only partly copied) or not a real video/audio file. Try the original file again."

The encoder found no readable streams in the file. Wait for the download or copy to finish, then pick the original file again. A file that is not really audio or video, such as a document renamed to .mp4, fails the same way.

"Original quality" is greyed out, or I got an .m4a instead of an MP3.

"Original quality" is off while "Normalize loudness", a "Volume boost" other than 0 dB, a Channels or Sample rate choice other than "Same as source", or "Mix all tracks" is on; the page explains this. Turn those off to use it. In that mode the extension comes from the source codec. For a particular format such as MP3, choose "Convert" and then the format.

It says "Failed to load the encoder. Check your internet connection and try again."

The encoder is about 30 MB and has to be downloaded from this site the first time you use a media tool. Check your connection and press "Extract audio" again. After one successful load it is kept in the browser's cache.

How it compares

"Original quality" and "Convert" answer different needs. Copying is exact and quick, but you get the source codec in whatever file type fits it. Converting is what you need for a particular format, a smaller file, mono, a different sample rate or a louder result, but each lossy re-encode of an already lossy track costs a little quality, and a higher bitrate cannot bring back detail. Compared with desktop software such as a video editor or the ffmpeg command line, this tool does one job, does not take a web link (download the video first), and is limited by your browser's memory, so very large files and long batches are better done on a computer with the software installed. It needs no install.