Yapdo Convo: Multilingual Conversational Speech Corpus with 30,000+ Hours

A conversational speech corpus by Liva AI (YC S25) containing over 30,000 hours of recordings from more than 8,000 speakers across 17 languages. The dataset card was last updated on 2026-05-21. The source audio was natively recorded with separate speaker channels, but samples are presented as combined conversations.

Use Cases

• Train automatic speech recognition (ASR) models based on natural, conversational audio.
• Develop speaker diarization systems based on natively recorded separate speaker channels.
• Build multilingual speech models based on coverage of 17 languages.
• Research conversational dynamics and natural speech patterns based on the dataset's emphasis on naturalness.

Strengths

• Contains over 30,000 hours of audio, with an estimated total of 50,000 hours.
• Includes recordings from more than 8,000 unique speakers.
• Covers 17 different languages.
• Source audio was natively recorded with separate speaker channels, which suggests potential for high-quality speaker separation tasks.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• A portion of the total estimated hours is still undergoing quality assurance.

Provenance

Source

Liva AI (YC S25)

Collection Method

Natively recorded conversational audio.

Freshness

Last updated 2026-05-21 02:40:42; freshness should be verified.