dictate.js is a small Javascript library for browser-based real-time speech recognition. It uses Recorderjs for audio capture, and a WebSocket connection to the Kaldi GStreamer server for speech recognition.

The API is modelled after Android's SpeechRecognizer. See the source code of lib/dictate.js and the usage in demos/demo.js.

The demos connect to the public services running on wss://bark.phon.ioc.ee that offer Estonian and English speech recognition.

The demos are available here. (Note that in order to use a wss-service the HTML-pages must be loaded over https.)

To run the demos on localhost, start a local HTTP server, e.g.:

python -m SimpleHTTPServer

and then open e.g. http://localhost:8000/demos/mob.html.

Known to work in

• Google Chrome 52.0 on Ubuntu desktop
• Google Chrome 45.0 on Android 5.1
• Firefox 37.0.1 on Ubuntu desktop (unstable: often seems to lose connection to microphone)
• Firefox 41.0 on Android 5.1 (unstable: often seems to lose connection to microphone)

• Dikteeri is an Estonian dictation demo by the Laboratory of Phonetics and Speech Technology (in Estonian). Uses dictate.js.
• Kõnele contains an Android front-end to the Kaldi GStreamer server