List your company

Linear correlation between human MOS and automatic metrics on TextrolSpeech (test)

0.749LCCSPAM (WavLM)
0.402680.492590.58250.67241Jan 9, 2026

Evaluation results

MethodLinks
SPAM (WavLM)TTS Model=ParlerTTS-mini2026.010.749
RA-CLAPTTS Model=Ground truth audio2026.010.726
SPAM (WavLM)TTS Model=Ground truth audio2026.010.721
SPAM (CLAP)TTS Model=ParlerTTS-mini2026.010.71
RA-CLAPTTS Model=ParlerTTS-mini2026.010.709
SPAM (CLAP)TTS Model=Ground truth audio2026.010.673
SPAM (WavLM)TTS Model=PromptTTS2026.010.665
SPAM (CLAP)TTS Model=ParlerTTS-large2026.010.653
SPAM (WavLM)TTS Model=ParlerTTS-large2026.010.648
RA-CLAPTTS Model=ParlerTTS-large2026.010.615
SPAM (CLAP)TTS Model=PromptTTS2026.010.584
SPAM (CLAP)TTS Model=PromptStyle2026.010.518
SPAM (WavLM)TTS Model=PromptStyle2026.010.515
RA-CLAPTTS Model=PromptStyle2026.010.503
RA-CLAPTTS Model=VoxInstruct2026.010.473
SPAM (WavLM)TTS Model=VoxInstruct2026.010.462
SPAM (CLAP)TTS Model=VoxInstruct2026.010.429
RA-CLAPTTS Model=PromptTTS2026.010.416