Skip to content

Benchmark

EmergentTTS-Eval Paralinguistics WER

Scores are percentages. Lower is better.

Models measured
9
Scores recorded
14
Sources
1
IQ weight
1.07

Score distribution

23.1%
Mean
20.9%
Median
6.2%
Std dev
18.3%
10th percentile
31.7%
90th percentile
17.7%
Min
17.7%40.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.07 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 data17.7%
2Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data18.3%
3Gemini 2.5 Flash TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data18.4%
4GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data19%
5GPT-4o mini TTSvoice: Alloy, baseline· EmergentTTS-Eval data20.6%
6GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data20.6%
7Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data20.9%
8GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data20.9%
9GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data21.2%
10GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data21.3%
11GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data23.2%
12Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data28.4%
13F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data31.7%
14Orpheus TTSvoice: Tara· EmergentTTS-Eval data40.9%

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