ZipVoice: Fast and High-Quality Zero-Shot Text-to-Speech with Flow Matching

Open paper
Benchmark Model Rank Results
text-to-speech-synthesis-on-librispeech-pc-test-cleanZipVoice-Distill (4 NFE)#1WER: 1.51UTMOS: 4.05Speaker Similarity: 0.657
text-to-speech-synthesis-on-librispeech-pc-test-cleanZipVoice#4WER: 1.64UTMOS: 3.98Speaker Similarity: 0.668
text-to-speech-synthesis-on-seed-tts-test-enZipVoice-Distill (8 NFE)#7WER: 1.62Speaker Similarity: 0.67
text-to-speech-synthesis-on-seed-tts-test-enZipVoice#9WER: 1.7Speaker Similarity: 0.697
text-to-speech-synthesis-on-seed-tts-test-zhZipVoice-Distill (8 NFE)#11CER: 1.34Speaker Similarity: 0.74
text-to-speech-synthesis-on-seed-tts-test-zhZipVoice#13CER: 1.4Speaker Similarity: 0.751

Rank counts only results with a code link.