Cantonese Spontaneous Dialogue
Spontaneous two-party conversations from native Cantonese speakers capture natural speaking rhythm and dialogue structure. Human-verified transcripts with utterance-level timestamps support production ASR, voice agents, and multilingual speech applications.
Dataset Overview
Commercial License
Cleared for production AI development and model training.
Native Speakers
Native Cantonese speakers covering Hong Kong Cantonese.
Human Verified
Human transcription with review, reviewed before delivery.
Production Ready
16 kHz / 16-bit PCM WAV with structured, validated metadata.
Hear the data before you license it.
A representative audio excerpt from this dataset. Full transcripts and speaker metadata are delivered with licensed data.
Audio Sample 1
25–34 · Hong Kong Cantonese
Built to a documented standard.
Every delivery matches the recording, audio, and annotation specification below.
Recording
Two-party conversational recordings in quiet indoor environments.
Audio
Annotation
- Human Transcript
- Speaker ID
- Timestamps
- JSONL / TXT
Utterance-level timestamps
Balanced, documented speaker coverage.
Native Cantonese speakers
Balanced
18–60
Hong Kong Cantonese
Verified at every stage of delivery.
What teams build with this dataset.
Automatic Speech Recognition
Train and benchmark ASR models against human-verified reference transcripts.
Voice Agents
Build voice assistants that hold up against real conversational speech.
Multilingual AI Products
Extend language coverage with consistent annotation across locales.
License this dataset for production AI.
Commercial licensing for production AI development. Terms are shared on request; evaluate representative samples before purchase.