Mozilla’s ‘Common Voice’ – a Crowdsourced Database For Voice Recognition Improvement

[ad_1]

I get lots of compliments about my deep Scottish accent but when it comes to voice apps such as Siri and Alexa, my voice is a liability more than an asset. When Apple and Amazon were making their voice assistants, they didn’t seem to have a Scottish accent consultant on speed-dial.

If voice recognition is supposed to be the way of the future – and supposedly that is where we are meant to be heading – then search engines and voice assistants need to start doing a better job of understanding difficult accents. We don’t all have public English schoolboy accents.

Mozilla (makers of the Firefox browser) are attempting to try and solve the problem by asking people to volunteer their voices to a database called “Common Voice“. By matching voices to set phrases, Mozilla is hoping that their database will eventually be an invaluable tool in the future to any app depending on voice recognition.

Signing Up For ‘Common Voice’

Before we go any further, let’s get the privacy issues out of the way first, as I can already hear them coming in.

While it is possible for you to sign up using your Firefox account, Github account, or Google account, you could instead sign up via email. Just open an anonymous throwaway email account and no-one will be any the wiser it is you speaking.

Once you have made an account and logged in, you will see your dashboard.

There are two sections – Speak and Listen. The Speak section is where you are asked to contribute your own voice. Listen is where you listen to other peoples voices and compare them to the phrase they had to speak. You would then indicate if it was correct.

Speak

To take part in the speaking section, you will need a functioning microphone that will pick up your voice clearly. You need to also give the Mozilla site access to that microphone.

When you click on the Speak section, you will be given a set of five random phrases to speak. The instructions are very clearly on the screen and easy to follow.

You simply click the microphone icon below and then read the sentence in your normal tone of voice. Don’t put on an unnatural accent or speak really slowly.

The whole point of the exercise is that the database learns to understand peoples accents the way they are normally spoken. Speaking in another way just makes the whole thing pointless.

When the first phrase has been done, you will see in the top-right corner the options to either playback the clip and to re-record the clip if you are not happy with it.

If you are happy with it, it moves onto the next one and continues till all five phrases are done. Then click the blue Submit button to send your clips to Mozilla for checking.

If you now go back to your dashboard, you will see that you have been credited for those phrases.

Then it’s a simple case of rinse and repeat. The more contributions and different accents Mozilla can get in their database, the more successful the project will be.

Listen

Once people have submitted their phrases to Mozilla, they need to be checked to make sure the voice clip matches the phrase. This is something which is also being crowdsourced out to volunteers.

Back on the dashboard, you will see the Listen section on the right. You can see from the Top Contributors section that some serious stats have been racked up.

To start validating voice phrases, go to the Listen section and you will be given a phrase and a play button.

Click on the play button (make sure your speakers are on!) and you will hear the recorded voice. If you decide the phrase was spoken correctly, click Yes. If there was mispronunciation, something else said, or any other error, click NO.

Conclusion

This is just one of the many projects which Mozilla has going at the moment (go to the main Mozilla website and click Projects at the top – it changes all the time). Crowdsourcing is a great way for worthy projects to happen and it is always worth donating your time to help build something useful for the future.



[ad_2]