How to Transcribe Video to Text: MP4, MOV, WebM and MKV Files

To transcribe a video to text you only need its audio track, so the job comes down to getting that audio into a speech-to-text engine and cleaning up the result. The quickest path is uploading the video file directly to a transcription service that accepts MP4, MOV, WebM and MKV; for very large files, extracting the audio first with a free tool such as ffmpeg or VLC makes the upload many times smaller. This guide covers both routes, plus timestamps, speaker labels, subtitles and export.

What "transcribing video" actually involves

A video file is a container (MP4, MOV, WebM, MKV) holding one or more streams: a video stream, one or more audio streams, and sometimes subtitles or chapters. Speech recognition only ever looks at the audio stream. That has three consequences:

• Video resolution, frame rate and codec make no difference to transcript quality. A 4K recording and a 360p recording of the same meeting transcribe identically.
• Everything that matters is in the audio: microphone distance, background noise, overlapping voices, compression.
• Uploading the full video is often wasteful. The audio track of an hour-long 1080p recording is typically a small fraction of the file size.

So the decision is mostly practical: is the file small enough to upload as it is, or should you pull the audio out first?

Formats: MP4, MOV, WebM, MKV and what is inside

If your tool lists the container as supported, try the direct upload first. If it refuses the file or stalls, extracting the audio (next section) resolves almost every case, because the output is a plain MP3 or M4A that every service accepts.

Option 1: upload the video directly

Most current transcription services accept video containers and extract the audio for you. The workflow in Audiotexta is representative:

1. Open the web app and drag the MP4, MOV, WebM or MKV file onto the upload area. If the video is on YouTube, paste the link instead (see YouTube video to text).
2. Choose the spoken language or leave auto-detect on, and turn on speaker labels if more than one person talks.
3. Wait for processing. On clear audio, automatic transcription normally finishes in a fraction of the recording's length.
4. Open the editor. Clicking a word plays the audio from that point and the current word is highlighted during playback, which makes corrections quick.
5. Export TXT or DOCX for a document, SRT or VTT if you want subtitles, or PDF for sharing.

Other tools that work this way include Otter, TurboScribe, Descript and the auto-caption features in YouTube Studio and Premiere Pro. Each has limits on file size and duration, usually per plan; check before uploading a three-hour recording.

Option 2: extract the audio first

Extracting audio is worth it when the file is large, the format is unusual, you are on a slow connection, or you want to keep a lightweight copy of the audio anyway.

ffmpeg (Windows, macOS, Linux)

ffmpeg is free and handles every container listed above. Two commands cover most needs.

Copy the existing audio stream without re-encoding (fast, no quality loss; works when the audio is AAC):

ffmpeg -i recording.mp4 -vn -c:a copy audio.m4a

Re-encode to a mono MP3 suited to speech (small file, universally accepted):

ffmpeg -i recording.mkv -vn -ac 1 -ar 16000 -b:a 64k audio.mp3

-vn drops the video, -ac 1 mixes to mono and -ar 16000 sets a 16 kHz sample rate, which is what most speech models use internally anyway. For WebM with Opus audio, the first command with an .ogg or .opus output extension copies the stream unchanged.

VLC (point and click)

Media → Convert/Save → add the file → Convert → choose an "Audio - MP3" profile → Start. VLC re-encodes, so it is slower than ffmpeg's stream copy, but it needs no command line.

QuickTime and online converters

On a Mac, QuickTime's File → Export As → Audio Only produces an M4A. Online converters work for small files, but avoid them for confidential recordings: you are uploading to a third party with unknown retention. HandBrake is a video converter and is not well suited to audio-only output.

Handling large files

Long recordings (webinars, all-day workshops, multi-camera edits) create three problems: upload time, service limits and your own patience. Options, in order of preference:

• Extract the audio. The single biggest win; it removes the video entirely.
• Reduce the audio bitrate. Speech is perfectly intelligible at 48–64 kbps mono; you do not need 320 kbps stereo for transcription.
• Split the recording. With ffmpeg, -ss 00:00:00 -t 01:00:00 cuts a one-hour segment; repeat for the rest. Many services also let you upload several files into one project.
• Trim silence and dead time. Cutting ten minutes of pre-meeting lobby chatter saves processing and keeps the transcript tidy.
• Use a link instead of a file. If the video already sits on YouTube or a cloud drive, pasting a link avoids the upload altogether.

Remember that on most services the limiting factor is duration, not megabytes. Check your plan's monthly minutes and per-file limit on the pricing page or the equivalent for your provider.

Timestamps, speaker labels and subtitles

Timestamps

Timestamps make a transcript navigable. Paragraph-level timestamps are fine for notes; word-level timestamps are needed if you intend to generate subtitles or sync the text to the video in an editor. Most services offer both, with the granularity chosen at export.

Speaker labels

Automatic speaker identification (diarization) separates "Speaker 1" from "Speaker 2" based on voice characteristics. It works well for two to four people with distinct voices and a decent microphone; it struggles with cross-talk, very similar voices and single-microphone room recordings. Rename the labels to real names once in the editor and the change applies throughout.

Subtitles

If the goal is captions rather than a document, export SRT or VTT straight from the transcript. SRT is the universal choice for video editors and media players; VTT is what HTML5 video and most web platforms expect. The subtitle generator guide covers line length, reading speed and burning captions into the video.

Editing the transcript efficiently

A raw automatic transcript is a draft. Typical clean-up takes a few minutes per ten minutes of clear audio and longer for difficult recordings. A checklist that works:

• Fix proper nouns first: people, companies, products, places. Use find-and-replace; errors repeat.
• Check numbers, dates and amounts against the audio. Speech models sometimes write "fifteen" as "fifty".
• Merge or split paragraphs so each covers one thought; automatic paragraphing is approximate.
• Remove filler ("um", "you know") if the transcript is for reading; keep it if it is for analysis.
• Mark anything genuinely inaudible as [inaudible] with a timestamp rather than guessing.
• Run the AI summary or chapters feature last, once the text is correct, so the summary is built on clean input.

Common video sources and their quirks

• Zoom, Teams and Google Meet recordings are MP4 (Zoom, Teams) or WebM (Meet). Separate audio tracks per participant, where available, give the best speaker separation. See the meeting transcription guide.
• Screen recordings (OBS, Loom, QuickTime, Xbox Game Bar) often record system audio and microphone on one track, so any notification sounds end up in the audio. Mute notifications before recording.
• Phone videos have the microphone far from the subject; wind and handling noise are common. Transcription still works, but expect more corrections.
• Lectures and webinars benefit more from chapters and summaries than from word-perfect text; generate those after the correction pass.
• Videos in a language you do not speak can be transcribed in the original language and translated afterwards. Translating the transcript is far more reliable than trying to transcribe directly into another language.

Privacy when transcribing video

Video often contains faces, names and sensitive discussions. If you upload to a cloud service, check where the data is processed, how long files are retained and whether you can delete them. For EU organisations this is a compliance question under the GDPR, not just a preference. Audiotexta lets you delete recordings and transcripts when you are done and does not use your content to train models. If you cannot send the audio anywhere, running Whisper locally is a legitimate alternative, covered in our Whisper vs cloud APIs comparison.

Whichever route you take, the pattern is the same: get the audio out of the container, transcribe it with timestamps, fix names, and export in the format the next step needs.