Benchmark RadarCN-NewsTTS Bench
CN-NewsTTS Bench v0.1 evaluates Chinese news TTS pronunciation of dense written forms (scores, model names, ranges, units, percentages, abbreviations) from raw text, using 800 public test records and a 992-target auto-evaluable subset with ASR-ensemble transcripts and an automatic target scorer.
- Released
- 2026-06-23
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Addresses the gap where TTS systems may preserve written strings while altering spoken meaning in Chinese news. Provides a reproducible target-level scoring contract for comparing pronunciation accuracy across systems without manual intervention.
Motivation
Chinese news text contains dense written forms such as scores, hyphenated model names, ranges, unit symbols, percentages, English abbreviations, and mixed Chinese-Latin-digit names.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.