Mozilla has created a voice recognition AI, dubbed Project Common Voice, and wants everyday people to stop by the project’s website regularly to help train the bot. The goal of the project is to provide a powerful voice recognition and natural language AI tool that’s open-source and easy to implement. While larger companies like Google have the money, resources, and manpower to create vast neural networks and comprehensive training programs in order to produce top of the line voice recognition AI, Mozilla wants to bring that kind of power to the masses. Project Open Voice is still in its infancy, but is fully open-source, and can be downloaded, used in projects, and even forked by anybody who would like to do so.
The programming part of the project is all being done on Mozilla’s end. The company is asking for users to head to the project page and contribute their voice by reading sentences. Creating a user profile with some demographic information that the program can use to help figure out the commonalities of your voice type and other factors can make your contribution even more effective. Additionally, you can listen to sentences read by others, and figure out whether the bot got the sentence right or not. These contributions seem simple, but with a large number of people helping out, it’s entirely possible to create a very sophisticated, well-trained AI this way.
Mozilla has a long history of championing open-source projects and bringing projects that would normally require vast resources and computing power out for the common good. Even with Mozilla’s experience, however, natural speech and voice recognition won’t be an easy target. Each different company out there has its own approach to voice recognition, and that’s simply because there are so many possible ways to go about it. Google, for example, relies heavily on machine learning and neural networking, as with most of its projects. There is less of a focus on working to amass large amounts of training material, opting instead to simply let that data come to them by way of Android and Chrome OS device owners making use of tools like Google Assistant. With that much data flowing naturally, Google’s researchers can instead focus on optimizing the model and building out the neural network. That’s just one example, of course; Mozilla’s approach is a novel one, and time will tell just how well it works and what the project is useful for in the future, should it take off as Mozilla wishes it to.