Common Voice Scripted Speech v27.0 and Spontaneous Speech v5.0 Released

Common Voice
1

The latest versions of Common Voice datasets, Scripted Speech v27.0 and Spontaneous Speech v5.0, are available for download on the Mozilla Data Collective! A giant thank you to all of our contributors for their hard work putting together these datasets.

Download the latest datasets

What’s new in this release:

  • Version 27.0 of Scripted Speech includes datasets for 295 languages. These datasets contain 32,349,999 voice clips, making approximately 42,593 hours of speech data available for developing and improving speech technology.

  • Version 5.0 of Spontaneous Speech includes datasets in 80 languages. These datasets contain 89,754 voice clips of free-form answers to questions, of which 302 hours have been transcribed and validated.

Since the last release, the Common Voice community has welcomed 1 language to Scripted Speech - Pa’O (blk) - and 6 languages to Spontaneous Speech - Chinese (China, zh-CN), Swahili (sw), Palauan (pau), Sundanese (su), Bengali (bn) and Shan (shn).

Contribute to your language.

Community Spotlights

Among many contributions, this release we would like to highlight outstanding contributions from the following communities.

  • Laz (lzz): The Laz dataset is the fastest growing dataset on Common Voice. Since the last release, the Laz community has increased the size of their Scripted Speech dataset by over 67%, adding more than 23,000 clips (34.3 hours).

  • Sindhi (sd): The Sindhi dataset has seen an increase of 27%, coming from a group of just over 30 contributors. They have contributed more than 14,000 clips (15.6 hours) in just 3 months.

  • Upper Sorbian (hsb): With just over 40 contributors, the Upper Sorbian community has added more than 1,000 clips (2.2 hours) to their Scripted Speech dataset, increasing the size of their dataset by more than 17%.

  • Since the last release, the Hausa community added variants to their dataset, with extra support from Hausa Community Facilitator Alkasim Yusuf Musa. Hausa speakers can now select which variant they speak in their profile.

“For us, Common Voice is not simply a platform for teaching AI to understand Laz. Perhaps for the first time, so many members of the Laz community have come together to write, read and record their own language. Building a comprehensive and robust Laz dataset has become a shared effort to preserve our cultural memory and heritage and, above all, a source of hope for the future of Laz.”

- İsmail Avcı Bucaklişi, President, Laz Institute

“The Foundation for the Sorbian People and the Language Centre Witaj are very happy to see the progress made in the Common Voice dataset for Upper Sorbian. We like to encourage people to keep contributing to both Upper and Lower Sorbian datasets, in order to support further development of speech technologies for these languages.”

- Daniel Sobe, The Foundation for the Sorbian People and the Language Centre Witaj

Become a Common Voice Community Facilitator

The Common Voice community is continuously growing, with extra help from our community facilitators. If you are interested in becoming a facilitator for your community, please apply.

3 Likes