Omnilingual ASR Corpus: Spontaneous Speech for 348 Under-Served Languages

Meta FAIR released the Omnilingual ASR Corpus in 2024, providing spontaneous speech recordings and transcriptions for 348 under-served languages. The collection was developed to facilitate the training of automatic speech recognition and spoken language identification models for low-resource linguistic contexts.

Use Cases

• Training automatic speech recognition models using the audio recordings and text transcriptions
• Developing language identification systems using the iso_639_3 and glottocode metadata
• Conducting linguistic analysis of spontaneous speech patterns across 348 distinct language codes

Strengths

• Covers 348 under-served languages
• Includes standardized glottocode and ISO 639-3 identifiers
• Features spontaneous speech recordings

Limitations

• Unknown total audio duration
• Potential transcription noise in low-resource contexts
• Lack of speaker-level metadata

Provenance

Source

Meta FAIR (Facebook AI Research)

Collection Method

Spontaneous speech recording and transcription

Freshness

Last updated November 2025.

Geography

Global