Mozilla is building an online voice library to train AI systems.
Called Project Common Voice, the aim is to collect voice samples from people around the world. Mozilla announced it alongside a slew of other initiatives aimed at ‘raising awareness for internet health’, including a live event that will tackle the FCC’s effort to end net neutrality and will feature Neko Case, Lauren Duca, Senator Al Franken, and other special guests.
With respect to Project Common Voice in particular, Mozilla said in a statement that “[v]oice-enabled devices represent the next major disruption, but access to databases is expensive and doesn’t include a diverse set of accents and languages.” Indeed, Google has also tried to tackle this kind of thing, for example by hiring Scottish people to record speech to train its technology on their unique accent.
The initiative thus represents both a public-interest and a business opportunity for Mozilla. As voice increasingly becomes the central user interface for mobile devices and the IoT, developers of AI platforms will need extensive bases on which to train their systems across various languages and accents, and Mozilla is clearly trying to stake out a high-profile position in this area. The company says its voice library will become publicly available later this year.
July 17, 2017 – by Alex Perala