Mozilla publishes a common voice maxi-update
Common Voice: more and more audio, more and more languages
Inside it contains a total of 7,226 hours of audio (5,591 of which are validated) in 54 different languages. A significant step forward compared to the 1,400 hours in 18 languages of the previous version published in the first months of last year. A project supported by the work of a community of volunteers who lent themselves to the analysis of the 5.5 million clips and to the addition of the metadata then used to instruct the recognition or speech synthesis algorithms, specifying for example the age of those who provided, sex and accent.Common Voice is designed to integrate with DeepSpeech, an open source suite for speech-to-tech and text-to-speech engines, as well as with developed by the Machine Learning Group of Mozilla.
Italian is among the five languages present in the Common Voice with over 5,000 different voices together with English, French, German and Spanish. Instead, there are seven with more than 500 hours of recordings: English, German, French, Catalan, Spanish, cabilo and kinyarwanda.
on the occasion, the software house has also made available the first datasets with a clip designed for a specific use: the pronunciation of the numbers “zero” to “nine”, the words “yes” and “no”, the commands “hey” and “Firefox” with approximately 120 hours of audio in 18 different languages. As already disclosed above will be useful also for the testing of the wakeword “Hey Firefox” .
Source: Mozilla
