ASCEND
Bilingual speakers in Hong Kong hold unscripted conversations across multiple turns. The released segments include Mandarin speech, English speech and switches between the two.
Visit official source (opens in a new tab)Research use
Evaluate mixed-language recognition and study how a model handles an English phrase inside a Mandarin turn.
Limitations
The corpus represents a small group of speakers in Hong Kong. Report results separately for mixed and monolingual segments.
License and access
- Dataset license
- CC BY-SA 4.0 (opens in a new tab)
- Access status
- Direct access
Download the official Parquet splits. Each includes audio, a transcription and metadata for language, topic and session.