Free actually means free here
Online transcription services charge per minute because they run the AI on their servers. This page flips the arrangement: it delivers the same class of speech model — OpenAI's Whisper — to your browser, where your own device does the work. That removes the three walls every "free" transcription service eventually hits: the per-minute meter, the upload of possibly confidential footage, and the signup gate.
How it works under the hood
Step one, the audio track is extracted and downsampled to 16 kHz mono with FFmpeg (WebAssembly) — the same engine behind our audio extractor. Step two, the Whisper model processes the audio in 30-second windows with overlapping strides, producing text with timestamps. Those timestamps are what let you export straight to SRT/VTT subtitles — making this a free subtitle generator as well as a transcriber. On machines with a modern GPU the model runs via WebGPU at several times real-time; on CPU, expect roughly the clip's own duration.
Getting the best accuracy
Pick the Accurate tier for anything you'll publish; set the language explicitly when you know it; and feed clean audio when possible — a lapel-mic recording transcribes far better than a phone across the room. Whisper handles natural speech, accents and technical vocabulary well, but it will still mishear homophones and names: budget a proofread pass, exactly as you would with any paid service. To fix subtitle timing afterwards, our subtitle sync fixer and format converter pick up where this tool leaves off.
FAQ
How do I transcribe a video to text for free, without signing up?
Choose a video or audio file above (or paste a direct media URL), pick the model quality, and click Transcribe. The audio is extracted and transcribed entirely on your device with the Whisper speech-recognition model — no account, no upload, no per-minute pricing, no watermark.
Is my video uploaded to a server?
No. Both steps — audio extraction (FFmpeg) and speech recognition (Whisper) — run in your browser via WebAssembly/WebGPU. The only downloads are the processing engines themselves; your media never leaves your device. That makes this tool usable for unreleased footage, client material and confidential recordings.
Which languages are supported?
Whisper is multilingual — it recognizes about 100 languages including English, Chinese, Spanish, Japanese, Korean, French and German. Auto-detect works well; explicitly choosing the language helps on noisy audio and short clips.
Can it generate subtitles, not just a transcript?
Yes — the transcription keeps timestamps, so you can download the result as SRT or VTT subtitle files ready to load into an editor or upload to YouTube, as well as plain text.
Why is the first run slow, and how accurate is it?
The speech model downloads once on first use (40–250 MB depending on quality tier) and is then cached by your browser. Accuracy: the Balanced tier handles clear speech very well; the Accurate tier approaches paid-service quality. Heavy accents, crosstalk and loud music will still produce errors — treat the output as a 95% draft that needs a proofread, which is true of every transcription service.
Can I transcribe from a YouTube or TikTok link?
No — platform links require server-side downloading, which free browser-based tools cannot do (and platform terms prohibit). The URL field works with direct media file links (ending in .mp4, .mp3, .wav…) whose server allows browser access. For platform videos you have rights to, download the file first, then upload it here.