Speech to Text

About this tool

Speech to Text offers two transcription modes. Live Mic uses your browser's built-in Web Speech API for real-time dictation in one of seven language options. There is no model download, but recognition is handled by the browser vendor's speech service — Chrome, for example, sends the microphone audio to Google's servers. Audio File mode runs OpenAI's Whisper Tiny model in your browser through Transformers.js: the file is decoded and transcribed locally and is not uploaded. The model (about 39 MB) and the library download from CDNs on first use, and the model files are kept in the browser's Cache Storage for later visits. File mode currently transcribes only the first 30 seconds of a file and ignores the rest, so cut longer recordings into clips first. The result is plain text you can edit and copy.

Why use it

• Live mode works instantly with no downloads via the Web Speech API for real-time dictation.

• File mode uses Whisper AI with no server upload

• Seven live-mode language options: English (US and UK), French, German, Spanish, Japanese and Chinese

• Whisper model files stay in browser Cache Storage after the first download

• Transcript can be copied with one click

How to use

1. Choose 'Live Mic' for real-time dictation or 'Audio File' for a recorded clip.
2. For live mic: pick a language, click 'Start Listening', and allow microphone access when prompted.
3. Speak and watch finished phrases appear in the transcript box; click Stop when you are done.
4. For a file: choose an audio file (clips up to 30 seconds are transcribed in full) and click 'Transcribe'.
5. Wait for the model download on first use, then for the transcript to appear.
6. Copy the transcript with the Copy button.

When it helps

• Dictating notes hands-free
• Converting short voicemails or audio memos to text
• Transcribing a 30-second clip from an interview or lecture
• Getting written text for a short clip in another language (transcribed, not translated)
• Accessibility use cases requiring spoken-to-written conversion

Examples

Tips

• Use Live Mic mode in a quiet room with a good headset microphone — Web Speech API accuracy is heavily dependent on input quality.
• For recorded audio, file mode (Whisper Tiny) produces dramatically better punctuation and capitalisation than the live API.
• The first file-mode run on a given browser downloads ~39 MB; trigger it on Wi-Fi if you're on a mobile data plan.
• File mode stops after 30 seconds of audio. Cut longer recordings into clips (the MP3 Cutter works) and transcribe each one.
• If live mode stops listening on its own, click Stop, then Start Listening again — the tool does not restart a session the browser has ended.
• For private recordings, use file mode: live mode in Chrome sends microphone audio to Google for recognition.

Frequently Asked Questions

Glossary

Whisper

An open-source automatic speech recognition system from OpenAI, trained on 680,000 hours of multilingual web audio and capable of recognising 99 languages.

Whisper Tiny

The smallest Whisper variant (39 MB, 39M parameters); fast enough to run in-browser via WebAssembly while still producing solid transcripts for clean speech.

Web Speech API

A browser API exposed by Chromium-based browsers (and Safari) that performs real-time speech recognition using the OS or vendor speech engine, often with cloud assist.

Transformers.js

A Hugging Face JavaScript library that runs Hugging Face transformer models (including Whisper) in the browser via ONNX Runtime Web.

ONNX Runtime Web

A WebAssembly build of Microsoft's ONNX Runtime that executes neural network models in the browser without a server.

Language detection

The automatic identification of the spoken language in an audio clip; Whisper performs this in its initial token prediction step.

Cache Storage

Browser storage for fetched files (the Cache API). Transformers.js saves the Whisper model files there so they download only once per browser profile.