Skip to content
Fan Zeng (Frank)
Back to resources

ContextASR-Bench

This benchmark pairs generated English and Mandarin speech with contextual cues for named entities across multiple domains.

Visit official source (opens in a new tab)

Research use

Compare no-context, domain-context and entity-context ASR. Test an agent’s context selection under fixed budgets and track both term recall and false insertions.

Limitations

The speech is synthesized. The benchmark measures contextual recognition under controlled conditions. Generalization to older patients requires recorded clinical speech.

License and access

Dataset license
MIT as listed in dataset release metadata (opens in a new tab)
Access status
Direct access

Use the publisher’s benchmark splits and read the dataset terms. Keep context hints and ground-truth transcripts separate during evaluation.

Additional sources