The set of data that Mozilla makes available to the public consists of some 400,000 recordings of 20,000 people.
Mozilla has been collecting voice samples through its Common Voice project . And, now, he has decided to release the first version of a large collection of voices that is known to be in the public domain.
In fact, its managers say that “it is already the second largest set of voice data available to the public we know.” And to this we should add that “people from all over the world are adding and validating new samples all the time”. Although at the moment the samples that are offered are in English, Mozilla assures that soon it will continue with “all the languages”.
The current repository has some 400,000 recordings belonging to 20,000 different people . In total it would be 500 hours of speech.
The objective of Mozilla with Common Voice is to help developers, manufacturers, companies and researchers when working with voice-enabled technology.
“Most of us only have access to a fairly limited collection of voice data,” they explain from Mozilla, adding that this is “an essential component for creating high-quality speech recognition engines . ” In addition, “this voice data can cost more than tens of thousands of dollars and its scale is insufficient to create a voice recognition at a level that people expect.”
“By providing this new set of public data,” Mozilla concludes, “we want to help overcome these barriers and facilitate the creation of new and better speech recognition systems.” Among them, his own Deep Speech.