Video Transcriber
Convert speech to text from any video or audio, powered by Groq Whisper, 50+ languages, any file size
Powered by Groq Whisper
50+ languages auto-detected · Audio extracted locally (any size video)
How it works:
1. Your browser extracts & compresses the audio locally (no size limit)
2. The small compressed audio is sent to Groq Whisper for transcription
3. Works with videos of any size: MP4, MOV, WebM, AVI, and audio files
Only the Compressed Audio Leaves Your Device, Not the Video
This is one of two tools on the site that isn't purely local, and it's worth being upfront about exactly what happens where. Your uploaded video never leaves your device: ffmpeg.wasm strips the video track and re-encodes the audio to a mono, 16kHz, 16kbps MP3 with the -vn flag, which is the same sample rate Whisper-based models expect natively and is compressed enough that even a two-hour recording usually lands around 7MB. Only that small compressed audio file is then POSTed to a server-side /api/transcribe endpoint, which forwards it to Groq's hosted Whisper API and returns the resulting text.
That's a meaningful difference from the rest of the tools on this site: everywhere else, nothing leaves your browser tab. Here, a compressed audio-only file does, because turning speech into text requires a trained speech recognition model that can't realistically run in a browser tab for a file of arbitrary length. In exchange, transcription handles video calls, lecture recordings, interviews, and podcasts recorded on video, in 50+ auto-detected languages, without you needing to manually extract audio first. If the compressed audio still comes out too large for the backend to accept, you'll get a clear error asking for a shorter clip rather than a silent failure.
How to Use
- 1
Upload a video or audio file
MP4, MOV, WebM, AVI, MP3, WAV, and similar formats are all accepted, regardless of length.
- 2
Click Transcribe
The two-step process starts: local extraction first, then the API call.
- 3
Audio is extracted and compressed in your browser
ffmpeg.wasm strips the video entirely and re-encodes the audio to mono 16kHz at 16kbps, a size built for speech, not music fidelity.
- 4
The compressed audio is sent for transcription
Only that small audio file, not your original video, is uploaded to the transcription API. This step typically takes 15-60 seconds.
- 5
Copy or download the transcript
The resulting text appears in a text box you can copy to clipboard or download as a .txt file.
Key Features
- →The only data that leaves your device is a compressed, speech-only audio file, roughly 16kbps mono, your original video file and its full audio quality never get uploaded.
- →Because the audio is aggressively compressed for speech rather than music, this tool is built for spoken content: meetings, lectures, interviews, and voiceovers, not for transcribing song lyrics from a music video.
- →Very long recordings still produce a small file at 16kbps (roughly 7MB per hour), so multi-hour meeting recordings are still practical to transcribe.
- →Clear speech with minimal background noise and music transcribes noticeably more accurately than noisy or overlapping audio.
- →The language is auto-detected, no need to specify it manually, and over 50 languages are supported.
- →If you get a 'file too large' error even after compression, trim the source video down with the Video Cut tool first, then transcribe the shorter clip.
Frequently Asked Questions
Does my video file get uploaded anywhere?
No, the full video never leaves your device. Only a small, compressed, audio-only file (mono, 16kHz, 16kbps, extracted locally via ffmpeg.wasm) is sent to the transcription API. This is different from most other EzyIMG tools, which never send anything anywhere.
Why does this tool send anything to a server when others don't?
Speech-to-text transcription requires a trained recognition model that isn't practical to run inside a browser tab for files of arbitrary length, so this tool routes the compressed audio through a server-side endpoint to Groq's hosted Whisper API, unlike the purely local ffmpeg.wasm tools elsewhere on the site.
Why is the audio compressed so heavily before it's sent?
16kbps mono at 16kHz is tuned for intelligible speech, not music fidelity, and keeps even long recordings small (about 7MB per hour), which is what makes transcribing a multi-hour recording practical without hitting upload size limits.
What happens if my video is extremely long?
Because the audio is compressed so aggressively, most videos up to a couple of hours still produce a small enough file to transcribe. If the compressed audio is still too large for the backend, you'll get an explicit error asking you to try a shorter clip.
How accurate is the transcription, and does it handle accents well?
It's powered by Groq's hosted Whisper model, which performs well on clear speech across 50+ auto-detected languages. Heavy accents, overlapping speakers, fast speech, or significant background noise can reduce accuracy somewhat, as with any speech recognition system.
Last updated: June 2026