audio & video to text
Drop a recording, get the words. Whisper AI runs right in your browser — no upload, no account, no minutes limit.
How it works
- Pick a file or record. Audio (MP3, M4A, WAV, OPUS voice notes) or video (MP4, WebM, MOV). Your browser extracts the audio and converts it to 16 kHz mono.
- The model loads once. The first time, your browser downloads the Whisper model (about 43–200 MB depending on model and device) and caches it.
- Transcription runs on your device. Long files are split at quiet moments and transcribed section by section, so you see text appear as it goes.
- Export. Copy the text, or download a .txt transcript or .srt / .vtt subtitles with timestamps.
Which model should I pick?
- Fast (whisper tiny) — smallest download, quickest. Good for clear English speech, voice memos and quick notes.
- Better (whisper base) — about twice the size and noticeably more accurate. Use it for Hindi and other Indian languages, accents, interviews and anything noisy.
On a laptop with Chrome or Edge, the tool uses your graphics chip through WebGPU, which is much faster. Without WebGPU it runs on the CPU, which works everywhere but is slower — a phone may need about as long as the recording itself.
Honest limits
The models that can run inside a browser are small. They're excellent for the price (free) and for privacy, but a paid cloud service with a much larger model will make fewer mistakes, especially with heavy accents, crosstalk or background music. This tool doesn't separate speakers and doesn't download videos from websites — use files you have and the right to transcribe.
Popular uses
- hindi audio to text
- video to text
- mp3 to text
- voice note to text
- subtitles for your youtube video
- srt subtitle generator
Need to tidy up the transcript? Try the AI rewriter to fix grammar or shorten it, or the word counter.
Frequently asked questions
Is it really free with no minute limits?
Yes. The speech-recognition model runs on your own device, so there are no server costs to pass on — no sign-up, no minutes counter, no watermark.
Is my audio uploaded?
No. Your file is decoded and transcribed inside your browser. The only thing downloaded is the AI model itself (once, from Hugging Face), which your browser then caches.
How accurate is it?
It uses OpenAI's open-source Whisper model in its tiny and base sizes. Clear speech comes out well, but it's less accurate than large paid cloud services — always proof-read names, numbers and technical terms.
Which languages are supported?
Whisper understands around 99 languages, including Hindi, Bengali, Marathi, Tamil, Telugu, Gujarati, Urdu, Spanish, French, Arabic, Japanese and Chinese. It can also translate speech in those languages straight into English text.
Why is it slow on my phone?
Your phone is doing all the AI work itself. Short clips are fine; for long recordings use a laptop with Chrome or Edge, which can use the graphics chip (WebGPU) and run several times faster.
Does it identify different speakers?
No — it writes what was said with timestamps, but doesn't label who said it.