How to pull the audio track from a video — free browser tool
- Step 1Open the extractor — Go to Video → MP3. FFmpeg-WASM loads once and runs offline thereafter — no account needed inside the free 50 MB / 30-minute limit.
- Step 2Drop your recording — Drag the interview, lecture, or meeting file (MP4, MOV, MKV, WebM — Zoom and Teams exports are usually MP4) onto the dropzone. It's read locally; the detected duration and channel count appear after FFmpeg probes it. Nothing uploads.
- Step 3Pick a voice-friendly bitrate — Open the Bitrate dropdown. For speech, choose 128 kbps · podcast (clean for any voice) or 64 kbps · voice (smallest, fine for a single mono speaker). The default is 192 kbps. This dropdown is the only control — there's no separate mono switch here, so a mono source already produces a small file.
- Step 4Pull the audio — Click the action button. FFmpeg discards the video track, decodes the speech audio, and re-encodes MP3 at your chosen
-b:a. A long webinar takes a few minutes because the WASM audio path is single-threaded, but it's all local. - Step 5Confirm length and clarity — The result panel shows the output size and engine. Check the duration matches the recording. Give it a quick listen — if there's hum, fan noise, or background chatter that will hurt transcription accuracy, note it for the next step.
- Step 6Prep for transcription or a feed — Save the MP3. For cleaner transcripts, run it through the AI noise reducer (RNNoise speech model) and then the loudness normalizer. To trim dead air at the start, use the silence stripper.
Bitrate choices for spoken-word audio
For speech, lower bitrates are fine and keep files small for transcription and feeds. These are the exact dropdown values. Sizes assume a one-hour recording.
| Dropdown label | FFmpeg flag | ~Size / hour | Use it for |
|---|---|---|---|
| 64 kbps · voice | -b:a 64k | ~29 MB | Single mono speaker, dictation, quick transcription uploads |
| 128 kbps · podcast | -b:a 128k | ~58 MB | Interviews, lectures, panels — clean for any speech, podcast-ready |
| 192 kbps · default | -b:a 192k | ~86 MB | Talks with music stings or higher-fidelity needs; tool default |
| 256 kbps | -b:a 256k | ~115 MB | Rarely needed for speech; use when there's significant music |
| 320 kbps · best | -b:a 320k | ~144 MB | Overkill for pure voice — reserve for music content |
Prepping pulled audio for transcription
Smaller, cleaner speech transcribes faster and more accurately. Pair the pull with these JAD audio tools.
| Goal | Tool | Why |
|---|---|---|
| Smallest file for upload | This tool at 64 kbps | An hour drops to ~29 MB, under most transcription size caps |
| Remove fan / room noise | AI noise reducer | RNNoise speech NN cleans steady noise that confuses ASR |
| Even out quiet/loud speakers | loudness normalizer | 2-pass EBU R128 brings both interview voices to one level |
| Cut long silences | silence stripper | silenceremove trims dead air, shortening transcription time |
| Trim to one segment | audio trimmer | Pull a single Q&A out of a long panel by timestamp |
Cookbook
Spoken-word extraction jobs — interviews, lectures, meetings — with the bitrate chosen and the local FFmpeg command. For speech, smaller is smarter.
Zoom interview (MP4) → small MP3 for transcription
A 50-minute recorded Zoom interview exports as MP4 with AAC stereo. You want the smallest decent file to feed a transcription service. 64 kbps mono-source keeps it tiny.
Source: interview_zoom.mp4 (AAC 48kHz, 50:00, 220 MB) Bitrate dropdown: 64 kbps · voice Tier: Pro (file > Free 50 MB) FFmpeg (in-browser): -i interview_zoom.mp4 -map_metadata 0 -map 0:a -map 0:v? \ -c:v copy -id3v2_version 3 -c:a libmp3lame -b:a 64k \ interview_zoom.mp3 Output: interview_zoom.mp3 (~24 MB, 50:00) — easy upload
University lecture (MKV) → 128 kbps MP3
A recorded lecture captured to MKV. You want a clean podcast-quality voice file for the commute. 128 kbps is the sweet spot for speech.
Source: lecture_week6.mkv (AAC stereo, 1:18:00) Bitrate dropdown: 128 kbps · podcast Output: lecture_week6.mp3 (~67 MB, 1:18:00) The video track is dropped; only the lecturer's audio remains. Follow with loudness-normalizer for consistent playback level.
Webinar replay (MP4) → audio podcast episode
You ran a webinar and want to repurpose the audio as a podcast episode. Pull the audio, then normalise loudness so it matches your other episodes.
Step 1: Video -> MP3, 128 kbps -> webinar_ep.mp3 Step 2: loudness-normalizer (Apple Podcasts / -16 LUFS) Step 3: audio-trimmer to cut the pre-roll countdown Result: a feed-ready episode, all processed locally.
Confidential meeting recording stays on-device
An internal all-hands recording must not leave the company network. Because extraction is 100% in-browser with no server API, pulling the audio is compliant by default.
Source: allhands_2026-06.mp4 (internal, confidential) Process: Video -> MP3 in browser (FFmpeg-WASM) Upload: none. No /api/v1 audio endpoint exists. The MP3 is written to your Downloads folder; the recording never touched a server.
Pull one answer from a long panel
A 2-hour panel video, but you only need the 4-minute answer from the keynote speaker. Pull the full audio first, then trim by timestamp.
Step 1: Video -> MP3 (whole panel) -> panel.mp3 Step 2: audio-trimmer, start 1:22:10 end 1:26:05 -> answer.mp3 The extractor pulls the whole track; trimming is step two.
Edge cases and what actually happens
64 kbps is fine for voice but thin for music
By design64 kbps is purpose-built for a single speaker — it keeps an hour to ~29 MB with clear speech. If the recording has music (intro stings, a performance), 64 kbps will sound rough on the music. Bump to 128 or 192 kbps when music is involved.
Recording exceeds free 30-minute / 50 MB limit
Tier limitMost interviews and lectures run past 30 minutes or 50 MB, hitting the Free cap (the two limits apply independently). Pro raises it to 120 minutes / 200 MB; Pro+Media to unlimited duration / 100 GB for all-day recordings.
Quiet interviewee vs loud interviewer
PreservedThe pull preserves source levels — it doesn't balance speakers. A close-mic interviewer can be much louder than a phone-distance guest. Run the result through the loudness normalizer (2-pass EBU R128) or speech leveler to even them out before transcription.
Background fan/AC noise hurts transcription accuracy
PreservedSteady room noise is extracted exactly as recorded — this tool doesn't denoise. Noisy audio lowers ASR accuracy. Follow with the AI noise reducer, which uses the RNNoise speech neural network (not a crude gate) to clean steady noise while keeping speech.
Stereo recording where speaker is only on one channel
First-track stereo keptSome setups record each mic to a separate channel. This tool keeps the source's channel layout (no -ac is set here), so you get a stereo MP3 with the speaker on one side. To collapse to mono, post-process with the channel splitter or a mono conversion step.
Teams/Zoom export with a separate screen-share track
First track usedSome meeting exports contain more than one audio stream. -map 0:a pulls the default/first track; there's no track-picker in this UI. If the wrong track comes out, the source has multiple audio streams — see the MKV→MP3 multi-track guide.
Video has no audio (slides-only capture)
RejectA screen capture of slides with the mic off has no audio stream — -map 0:a finds nothing and the job fails. Confirm the source plays with sound before pulling.
All-day recording exhausts browser memory
Resource limitEven within Pro+Media's 100 GB cap, a multi-hour, multi-gigabyte recording can run the tab out of memory on a modest device. Use a desktop browser, close other tabs, or split the recording first. Processing is fully local, so RAM is the practical limit.
Confidential audio must not leave the device
By designThere is no server-side audio API or runner — everything happens in the browser via FFmpeg-WASM. You cannot send a meeting recording to a JAD server for processing, so confidential audio stays local by construction. This is the privacy guarantee for sensitive recordings.
Duration shows slightly off on a VFR phone recording
MinorVariable-frame-rate captures can report a duration a fraction of a second off. The full audio is still extracted; only the displayed length may be marginally imprecise. The MP3 plays correctly end to end.
Frequently asked questions
What bitrate is best for pulling speech?
For a single speaker, 64 kbps · voice keeps the file tiny (~29 MB/hour). For interviews, lectures, and panels, 128 kbps · podcast is clean and podcast-ready. You only need 192 kbps or higher if the recording contains music. The default is 192 kbps.
Why pull audio instead of keeping the video?
An MP3 is a fraction of the size, and transcription services, audio editors, and podcast feed hosts all accept MP3 directly. Pulling the audio strips the heavy video track (-map 0:a), turning a multi-gigabyte capture into a manageable file.
Is my interview or meeting uploaded anywhere?
No. Extraction runs entirely in your browser via FFmpeg-WASM, and there's no server-side audio API at all. A confidential interview, deposition, or internal meeting recording is processed on your device and never leaves it.
Will the pulled MP3 be good enough for transcription?
Yes — speech transcribes well at 128 kbps, and even 64 kbps is fine for a clear single speaker. For best accuracy on noisy recordings, run the result through the AI noise reducer and loudness normalizer first.
Can I make the output mono to shrink it further?
This tool keeps the source channel layout — it doesn't have a mono switch. A mono source already produces a small file at 64 kbps. If you need to force-collapse a stereo recording to mono, do it as a follow-up with a channel/mono conversion step.
How long a recording can I pull from?
Free: up to 30 minutes / 50 MB. Pro: 120 minutes / 200 MB. Pro+Media and Developer: unlimited duration up to 100 GB — built for all-day recordings. Duration and size are checked separately.
Can I pull just one section of a long video?
The tool pulls the whole audio track. Extract the full MP3, then trim to the section you want with the audio trimmer using start and end timestamps. There's no time-range field in the extractor.
My interviewer and guest are at very different volumes — can this fix it?
Not on its own — the pull preserves source levels. Send the MP3 to the loudness normalizer (2-pass EBU R128) or the speech leveler to bring both voices to a consistent level.
Does it work on Zoom and Teams MP4 exports?
Yes — those are typically MP4 with AAC audio, which decodes cleanly to MP3. MKV, MOV, and WebM exports work too. If a meeting export has multiple audio tracks, the default track is used; see the MKV→MP3 guide.
There's a lot of background noise — does pulling clean it up?
No. The audio is extracted exactly as recorded. For cleanup, follow with the AI noise reducer, which uses the RNNoise speech neural network to remove steady noise like fans and AC while preserving the voice.
Is the pulled audio lossless?
No — MP3 is lossy and the source audio is re-encoded once with libmp3lame. For speech that's irrelevant (128 kbps is transparent for voice). If you need a lossless capture for archival, use the video-to-wav extractor.
What if the video has no audio track?
The job fails — -map 0:a finds no audio stream. Slides-only screen captures with the mic off are the usual culprit. Confirm the source plays with sound before pulling the audio.
Privacy first
Every JAD Audio tool runs entirely in your browser via FFmpeg (WebAssembly) and RNNoise. Your audio files never leave your device — verified by zero outbound network requests during processing.