Skip to content

Benchmark

EmergentTTS-Eval Questions Win-Rate

Scores are percentages. Higher is better.

Models measured
9
Scores recorded
13
Sources
1
IQ weight
0.56

Score distribution

52.1%
Mean
52.1%
Median
15.2%
Std dev
31.1%
10th percentile
74.8%
90th percentile
19.5%
Min
19.5%74.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.56 for speech models, measured over 7 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data74.8%
2Gemini 2.5 Flash TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data74.8%
3Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data65.7%
4GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data62.5%
5GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data55.5%
6GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data54.3%
7Speech 02 HDvoice: English_expressive_narrator· EmergentTTS-Eval data52.1%
8GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data49.6%
9GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data48.8%
10GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data45.7%
11Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data43.4%
12Orpheus TTSvoice: Tara· EmergentTTS-Eval data31.1%
13F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data19.5%

Data credits: EmergentTTS-Eval (Boson AI), Apache-2.0 (licence)