Skip to content

Benchmark

EmergentTTS-Eval Emotions WER

Scores are percentages. Lower is better.

Models measured
9
Scores recorded
14
Sources
1
IQ weight
0.73

Score distribution

1.5%
Mean
0.87%
Median
2.2%
Std dev
0.6%
10th percentile
1.8%
90th percentile
0.57%
Min
0.57%9.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.73 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
1Speech 02 HDvoice: English_expressive_narrator· EmergentTTS-Eval data0.57%
2Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data0.6%
3F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data0.7%
4Gemini 2.5 Flash TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data0.71%
5GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data0.71%
6Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data0.71%
7GPT-4o mini TTSvoice: Alloy, baseline· EmergentTTS-Eval data0.72%
8Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data0.87%
9GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data0.93%
10GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data0.95%
11GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data1%
12Orpheus TTSvoice: Tara· EmergentTTS-Eval data1.8%
13GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data1.8%
14GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data9.3%

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