Mozilla has recently updated the vocal sets of Common Voice, which now includes examples of pronunciation from more than 200 thousand people. The data has been made public as a treasure under the cc0 license, allowing it to be freely used. These sets are designed to be used in machine learning systems for developing models for speech recognition and synthesis.
This update has increased the volume of speech material in the collection from 31.8 to 32.6 thousand hours, with over 20 thousand hours having gone through the verification process. Additionally, the number of supported languages has increased from 129 to 131.
The English set had 93.9 thousand participants dictating 3587 hours of speech, with similar data for the Belarusian, Russian, Uzbek, and Ukrainian languages. The Belarusian set had 8444 participants and 1846 hours, the Russian set had 3296 participants and 278 hours, the Uzbek set had 2200 participants and 265 hours, and the Ukrainian set had 1104 participants and 114 hours.
The Common Voice project focuses on building a database of voice samples that represents a wide range of voices and speech styles. Users can contribute by recording phrases or evaluating existing data. This database, which includes diverse pronunciations of common phrases, is a valuable resource for machine learning systems and research projects.