AxonData/russian-speech-recognition-dataset

Russian Speech Recognition Dataset - 1,000+ Hours Call Center

1,000+ hours of real-world Russian call center audio with transcripts. Train speech recognition, sentiment analysis, and customer support AI models on authentic telephone conversations

Dataset Summary

Key Features

✅ 1,000+ hours of inbound & outbound calls
✅ 100% Russian telephone conversations
✅ Real-world audio - no synthetic data
✅ Full transcripts in Russian and in English

Full version of dataset is availible for commercial usage - leave a request on our website Axonlabs to purchase the dataset 💰

Audio Specifications

• Format: Single-channel (mono) telephone speech
• Sample rate: 8,000 Hz
• Quality: Real-world call center environment
• Annotations: Automatic (machine-generated) speaker role labels (Agent/Customer)

Real-world Russian call center recordings for:

• Speech Recognition & Audio Transcription (telephone quality, 8kHz)
• Sentiment Analysis from customer service dialogs
• Conversation Analytics (intent detection, topic classification)
• Speaker Diarization (Agent/Customer roles)
• Dialog Summarization for QA and CRM automation

Dataset Composition

Domains

• Customer Support: Technical issues, product questions, troubleshooting
• Sales: Product inquiries, upselling, cross-selling
• Billing/Account: Payment issues, account management, billing questions
• Finance: Account management, financial services
• Pharma: Healthcare-related customer service

Call Types

• Inbound: Customer-initiated calls
• Outbound: Agent-initiated calls

Speaker Roles

All recordings include role annotations:

• Agent: Call center representative
• Customer: End user