Uzbek YouTube Speech: 100K+ Transcribed Audio Segments via Gemini 2.0

Between 100,000 and 1,000,000 Uzbek language audio segments and transcriptions sourced from YouTube by openbank-uz in early 2026. The collection utilizes vocal isolation to separate speakers and Google's Gemini 2.0 Flash model for automated transcription.

Use Cases

• Automatic Speech Recognition (ASR) for Uzbek using audio segments and text transcriptions
• Text-to-Speech (TTS) synthesis for the Uzbek language
• Fine-tuning speech-to-text models on isolated vocal tracks

Strengths

• Contains between 100,000 and 1,000,000 records
• Uses Gemini 2.0 Flash for high-fidelity automated transcription
• Features vocal isolation to separate individual speakers from source videos

Limitations

• Transcriptions are AI-generated and lack human expert verification
• Restricted to non-commercial use via CC BY-NC 4.0 license
• Potential for residual background noise from YouTube source material

Provenance

Source

YouTube

Collection Method

Scraped from YouTube, processed with vocal isolation, and transcribed via Gemini 2.0 Flash

Freshness

Last updated March 2026.

Geography

Uzbekistan