Kaljurand / dictate.js

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

dictate.js

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.

API

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

Running the demos

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.:

python3 -m http.server 8081

and then open e.g. http://localhost:8081/demos/mob.html.

Browser support

Known to work in:

• Mozilla Firefox 73.0.1 on Ubuntu 19.10
• Chrome 80 on Android 10
• Opera 56.1 on Android 10
• various iOS and Windows devices (see #27)

Issues:

• Google Chrome 81.0 on Ubuntu 19.10 (WebSocket connection to 'wss://...' failed: Error in connection establishment: net::ERR_SSL_OBSOLETE_VERSION)

See also

• 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