Rohan Paul · @rohanpaul_ai · X·2026-09-09 03:23·47分钟前
AI 导读

Voice Arena 对 10 款 TTS 模型进行延迟实测并公开原始数据,Gradium 以 p50 231 ms 的最快首音频时间排名第一,且在 p25、p50、p75 上分布最紧凑。Rohan Paul 称该基准非自报数据,结果具有实际参考价值。

Rohan Paul@rohanpaul_ai
40AI 编辑部评分,满分 100
2026-09-09 03:23· 47分钟前
AI 导读

Voice Arena 对 10 款 TTS 模型进行延迟实测并公开原始数据,Gradium 以 p50 231 ms 的最快首音频时间排名第一,且在 p25、p50、p75 上分布最紧凑。Rohan Paul 称该基准非自报数据,结果具有实际参考价值。

I usually scroll past benchmark posts because they're self-reported. This one isn't: Voice Arena measures every model themselves and publishes the raw numbers.

Gradium earns fastest time-to-first-audio at 231 ms. This actually means something.

Congrats to the Gradium team!

GradiumVoice Arena benchmarked 10 TTS models on latency. Gradium lands #1: lowest p50 at 231 ms, and the tightest spread across p25, p50 and p75. This is one of the fi...

来源:Rohan Paul· x.com