Together AI releases SF Streets benchmark, revealing 39% street-name transcription error rate in speech models and cutting errors 60% with cross-lingual style transfer
Original title:How speech models fail where it matters the most and what to do about it
AI summary
Together AI released the SF Streets and US Streets benchmarks, testing 15 advanced ASR models on street-name transcription and finding an average error rate of 39%, with non-native English speakers 18 percentage points less accurate than native speakers (46% vs 64%).
Key Takeaways
- Together AI released the SF Streets and US Streets benchmarks
- testing 15 advanced ASR models on street-name transcription and finding an average error rate of 39%
- with non-native English speakers 18 percentage points less accurate than native speakers (46% vs…
Quick Facts
- Published
- Updated
- Source
- Together AI 研究与产品博客(RSS)
- Category
- Papers & Research
- AI score
- 63
原文用自建基准量化了 ASR 在街名转写上的短板,并给出可复用的合成数据改进方法,对语音部署有直接参考价值。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Together AI releases SF Streets benchmark, revealing 39% street-name transcription error rate in speech models and cutting errors 60% with cross-lingual style transfer. February 23, 2026. https://new.nat6.net/en/news/cmu1iytz90keerocnb7tqlmzs/