J. Pedro Ribeiro

Fun with Browser APIs: Speech Recognition

March 04, 2018

On this third post about browser APIs we're gonna talk about audio, or more specifically, speech. The Web Speech API is compromised of two interfaces, and today we're going to talk about one of them: Speech Recognition.

What is it about?

The Speech Recognition API takes a speech input (via microfone) and outputs a list of results, which including the transcript and the confidence level of that translation. It is customisable, letting you define what language you're using and what words (Grammar) you are interested in.

The confidence level (0-1) changes over time, meaning that as you speak, new results are output replacing the initial guess. When the system is happy with the results, a boolean flag isFinal comes true and it moves on to following parts of the speech.

Using the API to type as you speak

A simple speech to text application could be done with the following code:

const speech = new SpeechRecognition(); speech.lang = ‘en-US’; speech.onresult = (event) => { console.log(event.results[0][0].transcript } speech.start();

Grammar

On that example, more options could be applied to the speech variable, like grammar. Grammar must be written using the JSpeech Grammar Format, you can ream about it in the spec document.

Conclusion

With just a few lines we managed to translate speech into text, which then can be used by your application to do as you please. You could, for example, create a voice activated app by limiting the grammar and taking actions depending on which word was said.

A more complete example is available on this GitHub repo, where apart from a simple dictation, I've added a voice functionality to search for merchants on a website.


J. Pedro Ribeiro

Hello!
I’m a Brazilian front-end developer living in London. This website features some of my latest projects and my thoughts on anything web related.
You can find me on Twitter, Instagram, and LinkedIn.
Wanna talk? Send me a message.