Speech Data | 25,000 Hours of Customer Support Calls with Transcripts for ASR & Voice AI

This dataset features 25,000 hours of real, non-synthetic customer service speech data and call audio across industries such as retail, banking, entertainment, technology, and other consumer-facing sectors. It captures authentic customer-agent interactions, including natural speech patterns, hesitations, interruptions, clarifications, emotional responses, urgency signals, and escalation moments that are valuable for training speech-aware AI systems. The dataset is predominantly English, with approximately 96% English-language content and a strong focus on US English. Spanish, Hindi, and other languages may be available on request depending on project requirements and data availability. The dataset is designed for teams building ASR models, voice AI assistants, conversational agents, customer service automation, conversation intelligence tools, QA systems, and speech analytics models that need realistic support-call data rather than scripted or synthetic conversations. Dataset includes: • Real customer-agent call recordings • Predominantly English speech data, mostly US English • AI-generated transcripts with timestamps • Call-level metadata • PII-censored / redacted transcript data • Emotional and conversational signals such as urgency, interruptions, hesitation, repetition, raised voice, and escalation moments • Audio format options: OGG, FLAC, and WAV • 48 kHz audio available for OGG • 16 kHz audio available for FLAC and WAV • Dual-channel, speaker-separated audio options a