Skip to main content

Speech to text online

Transcribe speech in real time with your browser's built-in speech engine.

Free & unlimited
Mode
Language

Click to start recording

Transcript
Transcribed text will appear here.
ToolChamp never receives your audio. This uses your browser’s own speech engine — Chrome, Edge and Safari send the audio to their vendor’s speech service to transcribe it.

Dictate into your microphone and watch the text appear, then export it as plain text or as a subtitle file. This page drives the Web Speech API that is built into your browser, so there is no account, no upload form and no usage limit from this site — but it is important to understand where the recognition happens. Chrome, Edge and Safari send the captured audio to their own vendor's speech service to turn it into words. ToolChamp never receives your audio, and equally cannot promise it stays on your device, because the browser makes that call. Thirty-four language variants are selectable, continuous mode keeps listening through pauses, and each recognised phrase is timestamped so the transcript can be exported as SRT or WebVTT subtitles.

Key facts about Speech to text online

Key facts about Speech to text online
EngineThe browser's own Web Speech API (SpeechRecognition / webkitSpeechRecognition) — no model runs on this page
Where recognition happensChrome, Edge and Safari send the audio to their vendor's speech service. ToolChamp never receives it, and cannot claim it stays local
Languages34 variants including English (US, UK, Australia), Spanish (Spain, Mexico), Portuguese (Brazil, Portugal), Chinese (Simplified, Traditional), Japanese, Korean, Arabic, Hindi, Danish, Swedish, Norwegian and Finnish
Live textInterim results are on, so words appear as you speak and are corrected as the engine hears more context
Continuous modeOn by default — recognition keeps running through pauses instead of stopping at the first silence
TimestampsEvery finalised phrase is stored with a time, which is what makes SRT and VTT export possible
Export formatsPlain text (.txt), SubRip (.srt) and WebVTT (.vtt)
Speaker labelsManual: you switch the current speaker number yourself. There is no automatic diarization
Audio filesA dropped file is played through your speakers and re-captured by the microphone — it is not decoded and fed to the engine directly
File size cap500 MB, though the playback route makes long files impractical

What happens to your file

Be precise about this one. ToolChamp never receives your audio: no recording is uploaded to this site, nothing is stored on a server here, and the transcript lives only in your browser tab. But the recognition itself is performed by your browser's built-in speech engine, and in Chrome, Edge and Safari that engine sends the captured audio to the vendor's own cloud speech service. That is a property of the browser, not of this page, and no web tool built on the Web Speech API can avoid it. If your material is confidential, do not use a browser speech API for it — use offline transcription software on your own machine.

About this tool

  1. 1

    Pick your language

    Use the searchable picker to choose the variant, not just the language. English (UK) and English (US) produce noticeably different spelling and recognition of the same speech.

  2. 2

    Allow the microphone

    The browser asks once. Recognition cannot start without it, because the Web Speech API listens to a live input rather than reading a file.

  3. 3

    Start speaking

    Words appear as interim text and firm up as the engine hears more of the sentence. Continuous mode keeps it listening through natural pauses.

  4. 4

    Or play a recording into it

    Drop an audio file and use play-and-transcribe. The file plays through your speakers and the microphone hears it, so a quiet room and decent speakers matter a great deal.

  5. 5

    Mark speakers if you need to

    Turn on speaker labels and switch the current speaker number as the conversation changes hands. This is manual — nothing detects who is talking.

  6. 6

    Export

    Copy the text, or download TXT for a document, SRT for most video editors and platforms, or VTT for HTML5 video and web players.

Specs & compatibility
API requiredSpeechRecognition or webkitSpeechRecognition — present in Chrome, Edge and Safari; absent in Firefox by default
MicrophoneRequired. The browser asks for permission on first use
Languages34 selectable variants, searchable in the picker
ExportTXT always; SRT and VTT whenever timestamps were captured
Subtitle numberingSRT cues are numbered sequentially from the captured phrase boundaries
OfflineNo. If the browser's speech service is unreachable, recognition fails
File inputAccepts MP3, WAV, OGG, M4A, FLAC, AAC and WebM up to 500 MB, but transcribes them by playing them aloud
Accuracy factorsMicrophone quality, background noise, accent and how closely the selected language variant matches the speaker
  • Choose the right regional variant. Selecting English (US) for a British speaker measurably increases the error rate and changes the spelling of the output.
  • Speak in complete phrases. The engine uses context to correct itself, so an isolated word is recognised worse than the same word inside a sentence.
  • Punctuation is not dictated automatically in most engines — say the punctuation, or fix it afterwards in the transcript box.
  • Transcribing a file by playing it aloud works, but every bit of room noise and speaker colouration goes in with it. Clean the recording first with the noise reducer if you must use this route.
  • For subtitles, export VTT for web players and SRT for video editors and most social platforms. Both carry the same cue timings.
  • Firefox does not ship the Web Speech API enabled by default, so this page cannot work there. Use Chrome, Edge or Safari.
  • Do not use any browser speech API for confidential material. The audio goes to the browser vendor's service, which is outside this site's control.
  • Live dictation with interim results
  • 34 selectable language variants
  • Continuous mode that survives pauses
  • Per-phrase timestamps
  • TXT, SRT and WebVTT export
  • Manual speaker labelling
  • Play-and-transcribe for existing audio files
  • Copy the whole transcript to the clipboard
  • Dictating notes or a draft straight into text instead of typing
  • Producing a rough transcript of a meeting you recorded
  • Generating a first-pass subtitle file for a video you will then correct by hand
  • Capturing an interview in a language you can speak but type slowly in
  • Making a searchable text record of a lecture
  • Getting a quick written summary of a voice memo
Not necessarily, and it would be dishonest to say otherwise. This page uses the Web Speech API built into your browser, and Chrome, Edge and Safari implement it by sending the captured audio to their own cloud speech service for recognition. ToolChamp never receives your audio and stores nothing, but the browser's transmission is outside this page's control and cannot be disabled by any website. If the material is confidential, use offline transcription software installed on your own machine instead of any browser-based tool.
Firefox does not enable the Web Speech API's recognition half by default. This page checks for SpeechRecognition and its webkit-prefixed variant when it loads, and shows an unsupported notice when neither is present. There is no workaround from the page's side, because the recognition engine is a browser feature rather than something a website can ship. Chrome, Edge and Safari all support it. Firefox does support speech synthesis — the other half of the API — which is why some voice tools work there and this one does not.
In a limited way. Drop a file and use play-and-transcribe: the page plays it through your speakers while recognition listens on the microphone. That works, but it is acoustic re-capture, not direct decoding, so the result depends on your speakers, your room and your microphone, and it runs in real time — an hour of audio takes an hour. The Web Speech API has no way to accept a file directly. For serious file transcription, offline software that reads the file is a far better fit.
No. The speaker-labels feature is manual: you turn it on and switch the current speaker number yourself as the conversation moves between people, and the label is prefixed to each phrase from then on. Automatic speaker separation — diarization — requires clustering voice characteristics across the recording, which the Web Speech API does not expose and this page does not attempt. For a two-person interview it is quick enough to switch by hand as you go; for a panel it becomes impractical.
Every finalised phrase is stored with a timestamp, and the export menu turns those into either SubRip (.srt) or WebVTT (.vtt). SRT is what most video editors, YouTube and the major social platforms expect. VTT is the HTML5 standard and what you want for a video element on a web page. Both carry identical cue timings. Expect to correct the result: cue boundaries follow where the recognition engine decided a phrase ended, which is not always where a subtitle should break.
Good for clear speech in a quiet room with a decent microphone, and it degrades quickly away from that. The main variables are background noise, distance from the microphone, accent relative to the chosen language variant, and whether you speak in full phrases — the engine uses surrounding context to correct itself, so isolated words fare worse. Picking the right regional variant matters more than people expect. Treat any transcript as a first draft that needs reading through, not as a finished document.

Related tools

View all

Part of Audio tools that run in your browser

Updated

We use anonymous analytics to improve ToolChamp. No personal data is stored or sold. Privacy Policy