Extracting audio from video: lectures, podcasts and music without losing quality

A two-hour lecture filmed on a phone is seven gigabytes of video whose only useful part is the sound; the same goes for a recorded interview, a webinar or a gig filmed from the crowd. Getting the audio out is a small job, but there are two genuinely different ways to do it, and the wrong one either wastes an hour of encoding or quietly degrades a recording you cannot make again. By the end of this guide you will know which codec is inside your video, when you can lift it out untouched and when you must re-encode, how to trim the result to the second without another encode, and how to get a multi-hour file through an ordinary laptop with the tools on this site.

Two different jobs: copying the track and re-encoding it

A video file is a container — MP4, MOV, WebM, MKV — holding separate streams: one video, one or more audio, sometimes subtitles. The audio stream is already a finished, compressed bitstream in its own codec (AAC, Opus, MP3, AC-3), so “extracting” it can mean one of two things.

The stream copy is one command on any machine with FFmpeg: ffmpeg -i lecture.mp4 -vn -c:a copy lecture.m4a. -vn drops the video stream and -c:a copy forbids re-encoding, so a two-hour file finishes in the time it takes to read from disk and the result is exactly as good as the audio was inside the video. A transcode from a lossy source stacks two sets of losses: AAC at 128 kb/s re-encoded to MP3 at 128 kb/s is worse than either. Copy if you can; transcode only when the codec in the video is wrong for where it is going.

Which audio is inside which video

The container tells you the likely codec, and the codec tells you which file type can hold it without re-encoding.

To see exactly what you have, ask the file:

ffprobe -v error -select_streams a:0 \
  -show_entries stream=codec_name,sample_rate,channels,bit_rate \
  -of default=noprint_wrappers=1 lecture.mp4
# codec_name=aac
# sample_rate=48000
# channels=2
# bit_rate=128003

Without a terminal, VLC shows the same under Tools → Codec Information (Ctrl+J, Cmd+I on a Mac). If -select_streams a:0 prints nothing, the video has no audio stream at all — a common surprise with screen recordings made without a microphone. The containers and codecs guide covers the video side of the same question.

What the tools on this site actually do

Extract Audio

Extract Audio has no options: processing starts the moment a file lands, in FFmpeg compiled to WebAssembly inside the tab. It runs the equivalent of ffmpeg -i lecture.mp4 -vn lecture_output.mp4. Because no audio codec is named, FFmpeg re-encodes the sound with its built-in AAC encoder at that encoder’s default rate — 128 kb/s for stereo — so this is a transcode, not a stream copy, even when the source was already AAC. The output is an MP4 container holding a single audio stream, served as audio/mp4; that is precisely what an M4A is, so rename lecture_output.mp4 to lecture_output.m4a if a music library treats .mp4 as video. Several files at once process one after another into converted-videos.zip, and because this tool always takes the FFmpeg path the 2 GB input ceiling described below applies.

Convert to MP3 and its siblings

Convert to MP3 accepts audio and video alike, and every run is ffmpeg -i input -vn -c:a libmp3lame -q:a 2 output.mp3. The -vn is why a video can be dropped straight on it — extraction and conversion in one pass — and -q:a 2is LAME’s V2 preset: variable bitrate averaging around 190 kb/s, generally regarded as transparent for music. VBR spends fewer bits on silence and simple passages, so a lecture comes out smaller per minute than a song. No bitrate, channel or sample-rate settings are exposed. The result is lecture_output.mp3; a multi-file drop encodes sequentially into converted-mp3.zip. The sibling converters share the shell with different encoders: M4A (-c:a aac -b:a 192k), WAV (16-bit PCM), FLAC and OGG (-c:a libvorbis -q:a 5). All take a video as input; none copies the stream.

• Listen on a phone or in a podcast app — drop the video on Convert to MP3. One step, plays everywhere.
• Send to a transcription service or an editor — WAV if size is no object, M4A otherwise.
• Keep the original AAC bit-for-bit — the desktop -c:a copy command; Extract Audio is the nearest browser equivalent but it re-encodes.
• Only part of it is wanted — extract first, then cut with Trim Audio. Never the other way round.

Trimming without another encode

Trim Audiois the one tool here that does not re-encode. Load a file, find the in and out points in the player, press “Use current” beside Start and End (or type seconds in 0.1 s steps) and press “Trim & download”. Underneath it runs ffmpeg -i lecture_output.mp4 -ss 754.2 -to 3120 -c copy lecture_output_trimmed.mp4: packets are copied, not decoded; FFmpeg discards those before the start time and stops writing at the end time.

This is where audio behaves better than video. Compressed video can only be decoded from a keyframe, and keyframes sit one to ten seconds apart, so a copy-trim of a video snaps its start back to the previous keyframe. Audio frames are tiny and each is independently decodable: an AAC-LC frame is 1024 samples, 21.3 ms at 48 kHz or 23.2 ms at 44.1 kHz; an MP3 frame is 1152 samples, 26.1 ms at 44.1 kHz; an Opus frame is 20 ms. The cut lands within one frame of the time you typed.

Three details from the source. The output keeps the input’s extension — _trimmed. plus whatever you gave it — so trimming the extracted lecture_output.mp4 yields lecture_output_trimmed.mp4, still audio only. The trimmer accepts video files too, but -c copy copies every stream, so dropping the original lecture video on it produces a shorter video, not an audio file: extract first, then trim. And it is only a cut — no fade-in, fade-out, gain or normalisation — so put the in point in a pause, not mid-word.

Muting a video is the opposite job

Mute Videokeeps the picture and removes the audio stream entirely — gone from the file, not turned down. The options dialog opens with “Mute video” already ticked; Convert to, Resolution and Target file size are optional. For an MP4, MOV, MKV or WebM whose codec your browser can decode, the job runs through the device’s hardware path: the audio track is discarded and, as long as you leave resolution, size and compression alone, the video track is copied through untouched, which is why a mute of a large file finishes quickly. For AVI, FLV and WMV, or a codec the browser cannot decode, it falls back to FFmpeg with -an, which re-encodes the picture with libx264 and caps output at 1080p. Either way the result is name_output.ext with no audio stream — ffprobe shows a single Stream #0:0: Video line. If you want both halves, extract the audio first, then mute.

Bitrate and format, by what the recording is

The arithmetic is bitrate × 3600 ÷ 8 ÷ 1000. Mono halves the size, and a lecture is mono anyway: the camera’s single microphone is duplicated into two identical channels. The MP3 converter here is fixed at V2 and keeps the source’s channels, so for exactly 96 kb/s mono the command is ffmpeg -i lecture.mp4 -vn -ac 1 -b:a 96k lecture.mp3. Two conversions never help: a FLAC made from AAC is a large file with AAC quality, and MP3 320 from AAC 128 is AAC 128 in a bigger box. The audio formats guide goes through each codec in turn.

Two-hour recordings on a laptop

The in-browser FFmpeg loads the whole input into memory before it starts and cannot take files above roughly 2 GB — the options dialog says so: can't load files above ~2 GB. A two-hour 1080p phone recording is well past that: Apple’s own figures are 60 MB per minute for HEVC and 130 MB per minute for H.264 at 1080p30, so 7.2 GB or 15.6 GB. The desktop -c:a copy command handles any size in seconds. In the browser, shrink the video first with Compress Video: on its hardware path it re-encodes only the picture and carries the audio track across unchanged, and above 1.5 GiB on Chrome or Edge it streams the result straight to a file you choose, so the 2 GB ceiling does not apply. A 480p version of a lecture is a few hundred megabytes, and Extract Audio or Convert to MP3 takes it from there — the audio has still only been encoded once.

Encoding is CPU-bound and runs in the tab, so plug in and keep the tab in the foreground — browsers throttle background tabs. If a session runs to several hours, cut it into parts with the trimmer (0 to 3600, then 3600 to 7200) and name them so they sort: 2026-09-06-lecture-07-part2.mp3. Faster playback belongs to the player, not the file — VLC steps speed by 10% with ] and [, and every podcast app offers 1.25× to 2× with pitch correction — so never bake a speed change into the audio; it cannot be undone.

Troubleshooting

• Conversion failed. The file may be unsupported, corrupted, or contain no audio track. — from the audio converters. The usual cause is the last one: a screen recording made without a microphone, or a video someone already muted. FFmpeg itself would report Output file #0 does not contain any stream. There is nothing to extract.
• Conversion failed. The file may be unsupported, corrupted, or too large to process in this browser. — from Extract Audio or Mute Video. Over about 2 GB is the usual reason; compress first or use the command line.
• Could not trim this file. Its format may not allow fast trimming — try converting it first, or a different clip. — the copy trim failed; raw .aac (ADTS) streams and some .wma and .amr files do this. Convert to MP3 or M4A, then trim.
• This audio's format can't be previewed in the browser — enter the start and end times (in seconds) manually below. — the page’s <audio> element cannot decode that codec (Safari has no Ogg Vorbis playback). Read the times off VLC and type them; the trim still runs.
• Set an end time later than the start time. — the range must be at least 0.1 s.
• The extracted file shows a video icon, or the Music app refuses it. Rename .mp4 to .m4a; the contents are already an M4A.
• The trim came out as a video. The original video was dropped on Trim Audio, which copies every stream. Extract first.

One line on rights: extract audio from recordings you made or have permission to use — a lecture you attended is normally fine to keep for yourself; a concert bootleg or a streaming rip normally is not.