Multilingual Speech and Text Alignments from Bloom Library

Bloom-speech is a dataset of text-aligned speech audio sourced from bloomlibrary.org, containing over 50 languages including many low-resource ones. It is intended for training and testing speech-to-text or text-to-speech models. The dataset was created by sil-ai and was last updated in February 2023.

Use Cases

• Train automatic speech recognition models on multilingual audio-text pairs for over 50 languages, including low-resource ones like Languageboz and Languageceb.
• Develop text-to-speech systems using expert-generated text and speech alignments from the Bloom Library platform.
• Benchmark model performance across diverse language families represented by tags such as Languagebam, Languagechp, and Languagebjn.
• Analyze linguistic features or acoustic properties within specific language subsets like Languagecak or Languagebm.

Strengths

• Contains over 50 languages, providing broad multilingual coverage.
• Includes many low-resource languages, which are often underrepresented in speech datasets.
• Annotations and language data are expert-generated, suggesting a level of quality control.

Limitations

• The specific number of audio samples, rows, or total hours of speech is unknown.
• The dataset's geographic and temporal coverage for the included languages is unspecified.
• Without column details or sample data, the exact structure and features of the data are unclear.

Provenance

Source

bloomlibrary.org

Collection Method

Text-aligned speech data gathered from the Bloom Library platform.

Freshness

Last updated on 2023-02-15.