Multilingual Conversational Speech

[ OVERVIEW ]

A family of conversational speech corpora across 18+ locales, built for teams training multilingual ASR, voice agents, and dialogue systems. Each corpus features two-speaker dialogue captured in natural contexts: casual conversation, semi-scripted role-play, and domain-specific scenarios. Stereo recording with per-speaker channel isolation, word-level time alignment, full speaker diarization, and per-utterance emotion tagging across 18 categories. Licensed per-locale or as the full set.

[ KEY HIGHLIGHTS ]

• Stereo per-speaker channel isolation for clean diarization
• Word-level timestamp alignment across every utterance
• Per-utterance emotion labels with confidence scores
• 18 emotion categories including Joy, Confusion, Calmness, Doubt, and Determination
• Natural conversation dynamics: turn-taking, overlap, interruption
• Consented contributors with demographic metadata on file
• Licensed per-locale or as the full multi-locale family

[ TECHNICAL SPECIFICATIONS ]

Files

Stereo WAV, 44.1-48 kHz, 16-bit, with L/R speaker separation

Transcripts

JSON with word-level timestamps, speaker labels, and emotion annotations per utterance

Annotations

Speaker diarization · 18-category emotion tagging · optional custom layers per project

Licensing

Commercial training rights available · per-locale or full-family licensing · usage terms per dataset card