Mozilla has released an update to the Common Voice voice data sets, including pronunciation examples from around 200,000 people. The data is published as public domain (CC0). The proposed sets can be used in machine learning systems to build speech recognition and synthesis models.
Compared to the previous update, the volume of speech material in the collection has increased by 10% — from 18.2 to 20.2 thousand hours of speech. The number of supported languages has risen from 87 to 93. For 27 languages, over 100 hours of speech data has been accumulated, and for 9 — over 500 hours of speech data. For 9 languages, a proportion of female speech has also reached at least 45%.
Over 81,000 people participated in preparing materials in English, dictating 2,953 hours of speech (up from 79,000 participants and 2,886 hours). The Belarusian set includes 6,326 participants and 1,054 hours of speech material (up from 6,160 participants and 987 hours), the Russian language set has 2,585 participants and 201 hours (up from 2,452 participants and 193 hours), the Uzbek set includes 1,503 participants and 231 hours (up from 1,355 participants and 227 hours), and the Ukrainian language set has 696 participants and 79 hours (up from 684 participants and 76 hours).
The Common Voice project aims to organize collaborative efforts to accumulate a database of voice templates that reflects the full diversity of voices and speech styles. Users are invited to voice the phrases displayed on the screen or assess the quality of data contributed by other users. The accumulated database, containing recordings of various pronunciations of standard phrases in human speech, can be used without restrictions in machine learning systems and research projects.
Source: opennet.ru
