Speech to Text

Overview

Dictates speech into text using the browser's Web Speech API in eight languages, with a live interim transcript that finalises as you pause. Microphone permission is requested explicitly before recognition starts, errors are explained in plain language, and the finished transcript can be edited, copied or downloaded as a .txt file.

How It Works

Choose a language from the eight offered — US and UK English, Hindi, Spanish, French, German, Brazilian Portuguese and Japanese — then press Start listening. The tool first asks the browser for microphone access through getUserMedia, which surfaces the real permission prompt, and only then starts recognition with continuous and interim results enabled. Words appear in the transcript box immediately as live text that locks in at each pause; you can type corrections directly into the box. Press Stop listening to end, then Copy or Download .txt to take the transcript away.

Step-by-Step Usage Guide

1. Pick the language you will speak; switching is disabled while a session is running.
2. Press Start listening and grant microphone access when the browser prompts.
3. Speak naturally; interim words appear live and lock in as final text at each pause.
4. Stop, correct anything in the editable transcript, then Copy or Download .txt.

Technical Specifications & Standards

Recognition runs on webkitSpeechRecognition, the implementation the Web Speech API actually shipped in. The W3C specification for it never reached formal Recommendation status, and support is uneven: Chrome and Edge recognise reliably, Safari has prefixed support since 14.1, and Firefox does not expose the interface at all — the tool detects this and shows an explanatory panel instead of a dead button. Important consequence: on Chromium browsers recognition is not on-device. Per MDN, the browser uses a server-based recognition engine, so transcription requires an internet connection and your voice leaves the machine even though this site itself stores nothing. The tool warms the microphone with getUserMedia before starting recognition because that is the path that reliably triggers the permission prompt, then stops those tracks — it separates capture-blocked conditions (denied, no device, OS policy) from recognition-service failures, and maps the API's error codes (not-allowed, audio-capture, network, no-speech) to specific messages with a retry affordance. Interim results are accumulated separately from final ones; only finalised results are appended to the transcript, which is why the committed text does not jitter.

Targeted Use Cases

• Drafting emails, notes or documentation hands-free without installing a dictation app.
• Capturing quick voice memos as searchable text during a call or while reading.
• Testing how a Web Speech API integration behaves in your browser before writing your own code against it.
• Practising pronunciation in one of the eight supported languages and reading back what the recogniser heard.

Notes & Gotchas

• Use Chrome or Edge; Firefox has no SpeechRecognition support and the tool will tell you so instead of failing silently.
• Expect transcription to need a connection — Chromium browsers send audio to the vendor's speech service.
• Pause briefly at sentence boundaries; finalisation happens on pauses, and a single unbroken stream delays results.
• Check your browser's site permissions first if the mic seems stuck; a previously denied site can be reset there.

Frequently Asked Questions