Step-Audio EditX
Step-Audio EditX is a 3B audio model trained with reinforcement learning for iterative changes to emotion, speaking style and vocal effects. It also supports zero-shot text-to-speech.
Change the delivery of a recording
The model supports repeated edits to vocal delivery and pronunciation. Its card includes pinyin control for Chinese characters with multiple pronunciations.StepFun Step-Audio-EditX model card
Keep a clean reference between edits
Practical guidance. An editing workflow can accumulate changes across successive passes. Compare each result with the source recording, including words and timing that should stay fixed. Confirm weight-use terms before a commercial deployment.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Step-Audio-EditXStep-Audio-EditX-AWQ-4bit |
| Published updates | The release added Japanese and Korean in November 2025, then more vocal tags in January 2026.StepFun Step-Audio-EditX model card |
| License boundary | The card explicitly licenses code under Apache 2.0. It does not clearly extend that sentence to the weights.StepFun Step-Audio-EditX model card |
| Formats | The AWQ 4-bit repository is a quantized version.StepFun Step-Audio-EditX model card |
Related models
- Step-Audio TTS 3B
StepFun · Step-Audio
Step-Audio TTS 3B uses a dual-codebook language-model design for speech synthesis.
Launch date not confirmedDownloadable - StepAudio 2.5 TTS
StepFun · StepAudio 2.5
StepAudio 2.5 TTS uses overall context and inline directions to control a voice performance.
Launch date not confirmedAPI · Provider product