Skip to content

Benchmark

EmergentTTS-Eval Paralinguistics Win-Rate

Scores are percentages. Higher is better.

Models measured
9
Scores recorded
13
Sources
1
IQ weight
0.23

Score distribution

61.1%
Mean
68%
Median
26.6%
Std dev
19.8%
10th percentile
89.5%
90th percentile
5.9%
Min
5.9%91.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.23 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, strong-prompt· EmergentTTS-Eval data91.3%
2GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data89.5%
3Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data89.1%
4Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data84.3%
5GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data77.3%
6GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data71.6%
7GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data68%
8GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data63.4%
9GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data52.1%
10Orpheus TTSvoice: Tara· EmergentTTS-Eval data48.6%
11Speech 02 HDvoice: English_expressive_narrator· EmergentTTS-Eval data33.6%
12F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data19.8%
13Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data5.9%

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