concept · Web API · Speech

Punctuation Transcription

Speak naturally and compare what the engine produces with and without unspokenPunctuation. The left panel runs a recognition session without auto-punctuation; the right panel runs a separate session with it enabled. Both listen simultaneously.

Your browser does not support the Web Speech API (SpeechRecognition). Try Chrome 151+ on desktop.

Try saying: "I went to the store yesterday and I picked up some milk then I came home and made dinner it was really good"
(Say it naturally, without verbal punctuation — pause at sentence boundaries.)

unspokenPunctuation

Click "Start listening" to begin

unspokenPunctuation: false

Raw transcript will appear here…

unspokenPunctuation: true

Punctuated transcript will appear here…

Both panels listen to the same microphone input using two separate SpeechRecognition instances. The difference is a single property: recognition.unspokenPunctuation. On Chrome 151 with an on-device recognizer that supports auto-punctuation, you will see commas, periods, and question marks appear in the right panel without you having to say them.

the pattern

const raw   = new SpeechRecognition();
const punct = new SpeechRecognition();

// raw transcript — no auto-punctuation
raw.unspokenPunctuation = false;
raw.continuous          = true;
raw.interimResults      = true;

// punctuated transcript — engine infers pauses/sentence ends
punct.unspokenPunctuation = true;
punct.continuous           = true;
punct.interimResults       = true;

raw.onresult = e => {
  // e.results[i][0].transcript — plain words, no punctuation
};
punct.onresult = e => {
  // e.results[i][0].transcript — "Hello, how are you?"
};

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗

the pattern

see also

implementation reference