Espeech Balalaika: 100K-1M Annotated Russian Audiobook Speech Records

MTUCI's lab260 team released this Russian speech corpus in early 2026, containing between 100,000 and 1,000,000 records. The dataset consists of audiobook recordings filtered and annotated using the BALALAIKA pipeline to support advanced generative speech tasks.

Use Cases

• Training Text-to-Speech (TTS) models using the Russian transcriptions and audio pairs
• Fine-tuning Automatic Speech Recognition (ASR) systems for high-fidelity audiobook narration
• Developing speech generative models using BALALAIKA-annotated features

Strengths

• Contains between 100,000 and 1,000,000 records
• Annotated with the specialized BALALAIKA pipeline for generative quality
• Released under the permissive Apache-2.0 license

Limitations

• Restricted to the audiobook genre, which may not represent spontaneous conversational speech
• Limited to the Russian language only

Provenance

Source

MTUCI lab260, derived from ESpeech datasets

Collection Method

Filtered and annotated using the BALALAIKA pipeline

Freshness

Last updated February 2026.

Geography

Russia