Speech to Text — dictate live, or upload a recording

How dictation works here

1. 1

   Click the microphone

   Your browser asks for microphone permission once. Nothing installs, and nothing is sent to VexaScribe.

2. 2

   Talk normally

   Words appear as you speak. Say "period", "comma" or "new line" for punctuation, or add it afterwards in the box.

3. 3

   Copy or download

   Take the text as a copy-paste or a .txt file. Closing the tab discards it — nothing is stored.

Dictation runs for as long as you keep talking. Long silences can end a session in some browsers — if the text stops appearing, click the microphone again and carry on. Nothing already dictated is lost.

Where your voice actually goes

Most dictation tools say “private, in your browser” and leave it there. The honest answer is more specific, and it depends on which browser you are using.

If a recording is confidential enough that this matters, the safest option is the one that never touches a network: your operating system's own dictation for live speech, or a locally-run model for files. We would rather say that than pretend otherwise.

Accuracy: talking versus recording

Dictating usually beats transcribing, and the reason is not the software. When you dictate you are speaking clearly, close to a microphone, in a room you chose. A recording is whatever the room gave you.

What breaks dictation

• ●Speaking faster than you would to a person — the engine needs phrase boundaries to work with.
• ●Unusual proper nouns, which come back as the nearest common word.
• ●Background noise the browser cannot gate out, especially other voices.
• ●Long pauses, which end the session in some browsers mid-thought.

What breaks a recording

• ●Distance from the microphone — a phone across the table picks up as much room as speech.
• ●Crosstalk, where two people overlap and attribution lands on the wrong turn.
• ●Strong accents against a language with thinner training coverage.
• ●Technical vocabulary and acronyms, which sit outside everyday language.
• ●Music underneath speech, which shares the same frequency range as the voice.

For uploads we publish measured word error rates by language and audio condition rather than a single marketing number — clear audio, noisy interviews and heavily accented speech land in different places, and one figure would not survive contact with your recording. See how accurate transcription is for the full breakdown.

Find your file type

Every format below uploads here directly — no converting first. If you are not sure what you have, the descriptions should place it.

• M4AWhat the iPhone Voice Memos app records, and what most Apple devices hand you.
• MP3The default almost everywhere else — podcast downloads, dictaphones, most exports.
• WAVWhat a field recorder or a DAW produces. Uncompressed, large, best quality.
• OGGCommon from open-source tools and some messaging apps.
• MP4 and videoAny video file — the audio track is extracted for you.
• iPhone voice memosGetting the file off the phone is the awkward part; that page covers it.
• VoicemailCarrier-specific export steps, which differ more than you would expect.

Dictation for dyslexia and writing difficulty

Speaking and spelling are separate skills. For writers with dyslexia, dysgraphia, or anyone who finds a blank page harder than a conversation, dictation removes the step that causes the trouble — you say the sentence, and the spelling is handled.

It also changes what a first draft costs. Talking through an idea is faster than typing it for most people, and the resulting text tends to be more direct, because spoken sentences are shorter than written ones.

Two practical notes: dictate in whole sentences rather than word by word, since the engine uses surrounding words to choose between homophones. And expect to edit punctuation afterwards — say “period” and “comma” if you prefer, but many people find it faster to talk freely and punctuate at the end.

Speech to text, or text to speech?

They are opposite directions, and the names are close enough that people land on the wrong one constantly.