Skip to content

Benchmark

EmergentTTS-Eval Foreign Words WER

Scores are percentages. Lower is better.

Models measured
9
Scores recorded
14
Sources
1
IQ weight
0.42

Score distribution

15.6%
Mean
14.5%
Median
3.9%
Std dev
12.1%
10th percentile
22.3%
90th percentile
11.8%
Min
11.8%23.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.42 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 data11.8%
2GPT-4o mini TTSvoice: Alloy, strong-prompt· EmergentTTS-Eval data12.1%
3GPT-4o mini Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data12.7%
4Gemini 2.5 Flash TTSvoice: Zephyr· EmergentTTS-Eval data13%
5GPT-4o Audio Previewvoice: Ballad, strong-prompt· EmergentTTS-Eval data13.3%
6GPT-4o mini TTSvoice: Alloy, baseline· EmergentTTS-Eval data13.4%
7GPT-4o Audio Previewvoice: Alloy, strong-prompt· EmergentTTS-Eval data13.8%
8GPT-4o mini Audio Previewvoice: Alloy· EmergentTTS-Eval data14.5%
9Speech 02 HDvoice: English_expressive_narrator· EmergentTTS-Eval data14.6%
10GPT-4o Audio Previewvoice: Alloy· EmergentTTS-Eval data14.7%
11Gemini 2.5 Pro TTSvoice: Zephyr, strong-prompt· EmergentTTS-Eval data16.2%
12Kokoro 82Mvoice: af_heart· EmergentTTS-Eval data22.2%
13Orpheus TTSvoice: Tara· EmergentTTS-Eval data22.3%
14F5 TTSvoice: basic_ref_en· EmergentTTS-Eval data23.5%

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