Step-Audio-TTS-3B

Step-Audio-TTS-3B represents the industry's first Text-to-Speech (TTS) model trained on a large-scale synthetic dataset utilizing the LLM-Chat paradigm. It has achieved SOTA Character Error Rate (CER) results on the SEED TTS Eval benchmark. The model supports multiple languages, a variety of emotional expressions, and diverse voice style controls. Notably, Step-Audio-TTS-3B is also the first TTS model in the industry capable of generating RAP and Humming, marking a significant advancement in the field of speech synthesis.

This repository provides the model weights for StepAudio-TTS-3B, which is a dual-codebook trained LLM (Large Language Model) for text-to-speech synthesis. Additionally, it includes a vocoder trained using the dual-codebook approach, as well as a specialized vocoder specifically optimized for humming generation. These resources collectively enable high-quality speech synthesis and humming capabilities, leveraging the advanced dual-codebook training methodology.

Performance comparison of content consistency (CER/WER) between GLM-4-Voice and MinMo.

Model	test-zh	test-en
	CER (%) ↓	WER (%) ↓
GLM-4-Voice	2.19	2.91
MinMo	2.48	2.90
Step-Audio	1.53	2.71

Results of TTS Models on SEED Test Sets.

• StepAudio-TTS-3B-Single denotes dual-codebook backbone with single-codebook vocoder*

Model	test-zh		test-en
	CER (%) ↓	SS ↑	WER (%) ↓	SS ↑
FireRedTTS	1.51	0.630	3.82	0.460
MaskGCT	2.27	0.774	2.62	0.774
CosyVoice	3.63	0.775	4.29	0.699
CosyVoice 2	1.45	0.806	2.57	0.736
CosyVoice 2-S	1.45	0.812	2.38	0.743
Step-Audio-TTS-3B-Single	1.37	0.802	2.52	0.704
Step-Audio-TTS-3B	1.31	0.733	2.31	0.660
Step-Audio-TTS	1.17	0.73	2.0	0.660

Performance comparison of Dual-codebook Resynthesis with Cosyvoice.

Token	test-zh		test-en
	CER (%) ↓	SS ↑	WER (%) ↓	SS ↑
Groundtruth	0.972	-	2.156	-
CosyVoice	2.857	0.849	4.519	0.807
Step-Audio-TTS-3B	2.192	0.784	3.585	0.742

More information

For more information, please refer to our repository: Step-Audio.