T
traeai
Sign in

概念

WER

别名:词错误率

词错误率,衡量语音识别准确性

已跟踪 8 条高相关材料

TraeAI 观察

相关材料

已收录 8 条与 WER 相关的内容,按评分排序。

Hugging Face Blog 图标

Hugging Face 推出 Open TTS Leaderboard,通过客观指标解决 TTS 模型评估碎片化问题,评估时间从数周缩短至数小时。

入选理由:Open TTS Leaderboard 使用 WER/CER、RTFx、TTFA、SIM 等指标,评估时间从数周缩短至数小时

FeaturedArticle#TTS#评估#Hugging Face#开源英文
Hugging Face Blog 图标

The Open ASR Leaderboard Adds Its First Global South Language

Hugging Face Blog3070 字 (约 13 分钟)
85

Hugging Face推出全球首个覆盖印度语的开放ASR评估基准,通过多维度数据设计解决语音识别模型的系统性偏差问题。

入选理由:Monsoon评估集包含4,888名说话者,记录12个属性确保多样性

FeaturedArticle#ASR#Hugging Face#语音识别#评估基准#印度语英文
跨语言测试 WER (词错率)多个方向 上拿了第一名,错字最少

中文声说英文:3.19(CosyVoice2 是 17.10)
韩文声说英文:3.42(CosyVoice3 是 13.70)

Results released by Xiaohu show their speech conversion technology performs excellently in cross-language WER tests, with Chinese-to-English WER at only 3.19% and Korean-to-English WER at only 3.42%, significantly outperforming competitors like CosyVoice. This demonstrates significant breakthroughs in multilingual speech synthesis technology.

入选理由:小互技术中文转英文WER为3.19%,远优于CosyVoice2的17.10%

FeaturedTweet#Speech Recognition#Cross-Language Conversion#WER#Speech Synthesis中文
Adding Benchmaxxer Repellant to the Open ASR Leaderboard

Adding Benchmaxxer Repellant to the Open ASR Leaderboard

Hugging Face Blog1283 字 (约 6 分钟)
52

Hugging Face introduces private datasets to prevent models from over-optimizing on public ASR test sets, while keeping the public Average WER unchanged to preserve real-world performance measurement.

入选理由:引入私有数据集防止模型针对公开测试集过度优化(benchmaxxing)。

FeaturedArticle#ASR#Benchmark#Hugging Face#Benchmaxxing#WER英文

跨材料问答 · WER

回答基于:WER 相关 8 条材料
    0 / 500

    AI may generate inaccurate information. Please verify important content.