📝 Audio to Text Converter
Transcribe audio to text with Whisper AI running in your browser. MP3, M4A, WAV and more; copy text or download .txt and .srt. Audio is never uploaded.
The first run downloads the speech model from Hugging Face and keeps it in your browser cache; after that it works offline. Your audio is processed on this device and never uploaded.
What Audio to Text Converter Does
This audio to text converter transcribes speech with Whisper, an open speech-recognition model, running entirely inside your browser. Drop in an MP3, M4A, WAV, WebM or Ogg file — or the audio track of a video — and the transcript builds up on screen part by part. Copy the text, or download it as a plain .txt file or as .srt subtitles with timings.
Unlike most transcription sites, the audio is never uploaded. The file is decoded and processed on your device; the only thing downloaded is the model itself, from Hugging Face, the first time you use it. After that it is cached by your browser and the tool works even without a connection. There is no account and no per-minute allowance.
Two models are offered: a fast English-only model, and a larger multilingual model that detects the spoken language automatically or lets you choose it. Long recordings are split into pieces of up to 30 seconds — Whisper’s input length — at the quietest moment near each boundary, so words are rarely cut in half.
How to Use Audio to Text Converter
- Choose or drop an audio file (MP3, M4A, WAV, WebM or Ogg)
- Pick the English model, or the multilingual model and a language
- Press Transcribe — the model downloads once, then runs on your device
- Watch the text appear as each part is transcribed
- Copy the text or download it as .txt or .srt subtitles
Formula Used by Audio to Text Converter
How the audio is prepared
Decode → mix to mono → resample to 16,000 samples per second → split at quiet points into ≤ 30 s windows → transcribe each window
Worked example
A 5-minute stereo MP3 at 44.1 kHz.
- Decoded and resampled to 16 kHz mono: 4.8 million samples
- Split into about 11 windows of up to 28 seconds
- Each window transcribed with timestamps, offset by its start time
Result: One continuous transcript with timings for .srt subtitles.
Choosing a Model
| Model | Languages | Speed | Best for |
|---|---|---|---|
| Whisper tiny.en | English only | Fastest, smallest download | Voice memos, lectures, podcasts in English |
| Whisper base | Many languages, auto-detected | Slower, larger download | Non-English audio, mixed accents |
How to Read Your Result
Getting better results
Clear audio matters more than the model. Record close to the microphone, avoid background music, and let one person speak at a time. If the multilingual model guesses the wrong language on a short clip, choose the language yourself.
Proofreading
Automatic transcripts are drafts. Listen through once while reading, paying special attention to names, numbers and technical terms, which small models most often get wrong. Whisper can occasionally repeat a phrase or invent words during long silences; the timestamps view makes these easy to spot.
Limitations & Accuracy Notes
- Files are limited to 30 minutes to stay within browser memory; split longer recordings.
- Speakers are not identified — the transcript is one continuous text.
- The page can pause briefly while each part is processed, especially on phones.
- The first run needs an internet connection to download the model.