https://github.com/LevyGuy/tUnE.js

Web Speech recognition grammar POC for webkit using the Levenshtein distance algorithm
https://github.com/LevyGuy/tUnE.js

Last synced: 3 months ago
JSON representation

Web Speech recognition grammar POC for webkit using the Levenshtein distance algorithm

• Host: GitHub
• URL: https://github.com/LevyGuy/tUnE.js
• Owner: LevyGuy
• Created: 2014-02-15T23:01:14.000Z (over 10 years ago)
• Default Branch: master
• Last Pushed: 2014-03-18T06:51:14.000Z (over 10 years ago)
• Last Synced: 2024-04-04T03:40:11.672Z (7 months ago)
• Language: JavaScript
• Homepage:
• Size: 141 KB
• Stars: 31
• Watchers: 6
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• starred-awesome - tUnE.js - Web Speech recognition grammar POC for webkit using the Levenshtein distance algorithm (JavaScript)

README

        
# tUnE.js 

Speech recognition grammar POC for webkit using the Levenshtein distance algorithm

## About

Update:

I wrote an article with the help of Brian Rinaldi on [flippinawesome](http://flippinawesome.org/)

[here](http://flippinawesome.org/2014/03/10/improving-speech-recognition-in-the-browser/)

It was also on [HTML5 Weekly](http://html5weekly.com/issues/128)

After playing around a bit with the speech recognition API I've found out that the API fail to detect the correct sentence when encountering a person with bad English like myself :) ([here is a funny video to demonstrate that](http://youtu.be/5FFRoYhTJQQ)).

This could be a problem when the app is waiting for the user input and the API fail to deliver.

The correct way to tackle this would be using the speech recognition grammar API, however I could not get it to work for some reason (update: see the reason [here](https://groups.google.com/a/chromium.org/forum/#!msg/chromium-html5/Rh2a-PW90lU/EJkYomTnp8kJ)).

So I've decided to write a little POC on a speech recognition grammar API using the [Levenshtein distance algorithm](http://en.wikipedia.org/wiki/Levenshtein_distance). 

The app will pass an array of strings and the admissible distance between each string and the speech API results. Thus even if the speech recognition API failed to deliver the accurate phrase we can still say that the distance between our expected input and the API results is a valid distance and execute any relevant callback.

## Examples

I've created a little “cube test” to test the results.

The first test can be viewed here: http://youtu.be/SrXxLkWRf8A

it compares the results between 

1. a cube with the grammar plugin with no intrim results

2. a cube with the grammar plugin with intrim results

The second test is a little plugin for cnn. You can view the demo here: http://youtu.be/2nnhcHqt4Vk

the plugin will collect all the links on the page and will pass them to the grammar plugin. Once we get a valid input the plugin will click the article link.

## How to use

```javascript

tUnE({

    prefix: "ok google",    // Prefix to start listening

    prefixConfidence: 7,    // The distance for the prefix

    interimResults: false,  // Check against Intermediate results (I'm still testing that).

    prefixCallback: function (){

        // Listening

    },

    doneOperation: function (){

        // Stop

    },

    confidence: 4,          // Global confidence. default is 5.

    operations: [

        {

            say: ["move left", "move right"],       // Commands

            confidence: 5,  // Private confidence

            exec: function(phrase, confidence, event){

                console.log("moving", phrase, confidence, event); // We have results

            }

        },

        {

            say: "i'm done",    // We can have an array or a string

            exec: function(){

                console.log("I'm done");

                tUnE.reInit();  // Re-init the listener

            }

        }                            

    ]

});

```