Deepspeech Balalaika: Russian Speech Corpus for Generative Tasks

A curated Russian speech dataset for advanced speech generative tasks. The corpus was filtered and annotated by the lab260 team at MTUCI using the BALALAIKA pipeline. It includes genres such as podcasts, public speech, YouTube content, audiobooks, phone calls, and TTS.

Use Cases

• Training Russian speech recognition models based on the annotated audio corpus.
• Developing text-to-speech systems using the diverse genres like audiobooks and podcasts.
• Fine-tuning generative speech models on the filtered and high-quality Russian audio data.

Strengths

• Dataset is described as high-quality and meticulously filtered.
• Covers multiple genres including podcasts, public speech, YouTube, audiobooks, phone calls, and TTS.
• Annotated using a specific pipeline (BALALAIKA) by a named team (lab260 at MTUCI).

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.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Deepspeech (GitHub link)

Collection Method

Curated, filtered, and annotated by lab260 at MTUCI using the BALALAIKA pipeline.

Freshness

Last updated 2026-01-29 15:40:21; freshness should be verified.