Skip to content

Benchmark

EmergentTTS-Eval Complex Pronunciation WER

Scores are percentages. Lower is better.

Models measured
9
Scores recorded
14
Sources
1
IQ weight
0.75

Score distribution

34.4%
Mean
34.8%
Median
3.9%
Std dev
29.9%
10th percentile
41%
90th percentile
28.8%
Min
28.8%42.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.75 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 data28.8%
2GPT-4o mini TTSvoice: Alloy, baseline· EmergentTTS-Eval data29.9%
3Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data30.1%
4GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data31.6%
5Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data31.9%
6GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data32.3%
7Gemini 2.5 Flash TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data33.6%
8Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data34.8%
9GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data35.3%
10GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data35.9%
11GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data36.9%
12GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data37.1%
13Orpheus TTSvoice: Tara· EmergentTTS-Eval data41%
14F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data42.4%

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