Datatang

Dec 22, 2021

3 min read

Build Multilingual Speech Recognition System with High-quality Training Data

With the boost of “One Belt, One Road” policy and AI and cloud computing technology, more and more Chinese tech companies has gone global. However, for some AI companies, the road to go abroad still faces many problems. Language is one of the problems, smart products that can recognize local languages are a powerful tool to open up the local market.

Due to the differences between languages, AI manufacturers need to build models separately according to the characteristics of each language. In order to ensure the effect of speech recognition system, it is necessary to use high-quality training data of different languages to train the model. However, the lack of high-quality, multilingual training data becomes a major problem for speech recognition system.

As a world’s leading AI data services provider, Datatang has developed a series of speech datasets in more than 30 languages. All the data is recorded by native speakers with signed authorization agreements and data quality exceeds the data industry standard.

German Speech Data

Nearly 3,000 hours German speech data, the data is recorded by German native speakers. The recorded text is designed by linguistic experts, covering generic, interactive, on-board, home and other categories.

French Speech Data

Nearly 1,800 hours French speech data, the data is recorded by native speakers from France, Canada and Africa. The recording text is designed by linguistic experts, which covers general interactive, in-car and home category.

Spanish Speech Data

Nearly 3,000 hours Spanish speech data, the data is recorded by native speakers from Spain, Mexico, Columbia, Venezuela etc. The recorded script is designed by linguists and cover a wide range of topics including generic, interactive, in-vehicle and home.

Korean Speech Data

Nearly 2,000 hours Korean speech data, recorded by Korean native speakers. The recordings include economics, entertainment, news, oral, figure, letter.

Japanese Speech Data

Nearly 1,000 hours Japanese speech data, the data is recorded by native Japanese speakers. The recorded script is designed by linguists and cover a wide range of topics including generic, interactive, in-vehicle and home.

Hindi Speech Data

Nearly 1,500 hours Hindi speech data, recorded by Indian native speakers. The accent is authentic. The recording text is designed by language experts and covers general, interactive, car, home and other categories.

If the above data cannot meet the needs of your current research, Datatang also provides data customization services for specific groups of people, specific scenarios, and specific languages to meet customers’ diversified data needs.

End

If you need data services, please feel free to contact us: info@datatang.com