Mozilla has recently released an update to the voice data repository Common Voice, which now includes examples of pronunciation from over 200 thousand individuals. The data has been made publicly available under the cc0 license. This dataset can be utilized in machine learning systems to develop speech recognition and synthesis models. The collection now boasts an increase in speech material from 31.1 to 31.8 thousand hours, with 20.8 thousand hours having undergone verification. Additionally, the number of supported languages has risen from 124 to 129, with the inclusion of languages from African tribes such as spit, kidavida, Doluo, and Tsvana.
A total of 93.3 thousand individuals participated in providing 3554 hours of speech data for English materials, up from 92.3 thousand participants and 3508 hours in the previous update. For the Belarusian language, 8400 participants contributed to 1815 hours of speech material, compared to 8291 participants and 1766 hours in the previous release. Russian language data now includes 3241 participants and 277 hours, up from 3206 participants and 274 hours, while Uzbek language data has 2189 participants contributing 265 hours, compared to 2170 participants and 264 hours. Ukrainian language data saw 1091 participants and 113 hours, up from 1075 participants and 112 hours.
The Common Voice project aims to collaborate in building a diverse voice template database that captures a wide range of voices and speech styles. Users are encouraged to voice phrases displayed on the screen or assess the quality of data provided by others. This open database, containing recordings of various pronunciations of common phrases in human speech, can be utilized in machine learning systems and research initiatives.