Step-Audio-EditX

🔥🔥🔥 News!!！

• Jan 29, 2026:

  • 🧩 New Model Release:
    • Better performance, with an overall improvement of over 4%.
    • More paralinguistic tags have been added, including exhale, snort, inhale, chuckle, clears throat, giggle.
    • Welcome to try out at StepFun Audio Studio
  • 💻 We release the SFT, DPO and GRPO training code.
  • 🌟 Training and inference for vLLM are now supported. Thanks to the vLLM team!

• Nov 28, 2025: 🚀 New Model Release: Now supporting Japanese and Korean languages.

• Nov 23, 2025: 📊 Step-Audio-Edit-Benchmark Released!

• Nov 19, 2025: ⚙️ We release a new version of our model, which supports polyphonic pronunciation control and improves the performance of emotion, speaking style, and paralinguistic editing.

• Nov 12, 2025: 📦 We release the optimized inference code and model weights of Step-Audio-EditX (HuggingFace; ModelScope) and Step-Audio-Tokenizer(HuggingFace; ModelScope)

• Nov 07, 2025: ✨ Demo Page ; 🎮 HF Space Playground

• Nov 06, 2025: 👋 We release the technical report of Step-Audio-EditX.

Introduction

We are open-sourcing Step-Audio-EditX, a powerful 3B-parameter LLM-based Reinforcement Learning audio model specialized in expressive and iterative audio editing. It excels at editing emotion, speaking style, and paralinguistics, and also features robust zero-shot text-to-speech (TTS) capabilities.

Wechat developer group

📑 Open-source Plan

• [x] Inference Code
• [x] Online demo (Gradio)
• [x] Step-Audio-Edit-Benchmark
• [x] Model Checkpoints
  • [x] Step-Audio-Tokenizer
  • [x] Step-Audio-EditX
  • [x] Step-Audio-EditX-Int4
• [ ] Training Code
  • [x] SFT training
  • [x] DPO training
  • [x] GRPO training
  • [ ] PPO training
• [ ] ⏳ Feature Support Plan
  • [ ] Editing
    • [x] Polyphone pronunciation control
    • [x] More paralinguistic tags ([Cough, Crying, Stress, etc.])
    • [ ] Filler word removal
  • [ ] Other Languages
    • [x] Japanese, Korean
    • [ ] Arabic, French, Russian, Spanish, etc.

Features

• Zero-Shot TTS

  • Excellent zero-shot TTS cloning for Mandarin, English, Sichuanese, and Cantonese.
  • To use dialect or other languages, just add a [Sichuanese] / [Cantonese] / [Japanese] / [Korean] tag before your text.
  • 🔥 Polyphone pronunciation control, all you need to do is replace the polyphonic characters with pinyin.
    • [我也想过过过儿过过的生活] -> [我也想guo4guo4guo1儿guo4guo4的生活]

• Emotion and Speaking Style Editing

  • Remarkably effective iterative control over emotions and styles, supporting dozens of options for editing.
    • Emotion Editing : [ Angry, Happy, Sad, Excited, Fearful, Surprised, Disgusted, etc. ]
    • Speaking Style Editing: [ Act_coy, Older, Child, Whisper, Serious, Generous, Exaggerated, etc.]
    • Editing with more emotion and more speaking styles is on the way. Get Ready! 🚀

• Paralinguistic Editing

  • Precise control over 10 types of paralinguistic features for more natural, human-like, and expressive synthetic audio.
  • Supporting Tags:
    • [ Breathing, Laughter, Surprise-oh, Confirmation-en, Uhm, Surprise-ah, Surprise-wa, Sigh, Question-ei, Dissatisfaction-hnn ]

• Available Tags

emotion

happy

Expressing happiness

angry

Expressing anger

sad

Expressing sadness

fear

Expressing fear

surprised

Expressing surprise

confusion

Expressing confusion

empathy

Expressing empathy and understanding

embarrass

Expressing embarrassment

excited

Expressing excitement and enthusiasm

depressed

Expressing a depressed or discouraged mood

admiration

Expressing admiration or respect

coldness

Expressing coldness and indifference

disgusted

Expressing disgust or aversion

humour

Expressing humor or playfulness

speaking style

serious

Speaking in a serious or solemn manner

arrogant

Speaking in an arrogant manner

child

Speaking in a childlike manner

older

Speaking in an elderly-sounding manner

girl

Speaking in a light, youthful feminine manner

pure

Speaking in a pure, innocent manner

sister

Speaking in a mature, confident feminine manner

sweet

Speaking in a sweet, lovely manner

exaggerated

Speaking in an exaggerated, dramatic manner

ethereal

Speaking in a soft, airy, dreamy manner

whisper

Speaking in a whispering, very soft manner

generous

Speaking in a hearty, outgoing, and straight-talking manner

recite

Speaking in a clear, well-paced, poetry-reading manner

act_coy

Speaking in a sweet, playful, and endearing manner

warm

Speaking in a warm, friendly manner

shy

Speaking in a shy, timid manner

comfort

Speaking in a comforting, reassuring manner

authority

Speaking in an authoritative, commanding manner

chat

Speaking in a casual, conversational manner

radio

Speaking in a radio-broadcast manner

soulful

Speaking in a heartfelt, deeply emotional manner

gentle

Speaking in a gentle, soft manner

story

Speaking in a narrative, audiobook-style manner

vivid

Speaking in a lively, expressive manner

program

Speaking in a show-host/presenter manner

news

Speaking in a news broadcasting manner

advertising

Speaking in a polished, high-end commercial voiceover manner

roar

Speaking in a loud, deep, roaring manner

murmur

Speaking in a quiet, low manner

shout

Speaking in a loud, sharp, shouting manner

deeply

Speaking in a deep and low-pitched tone

loudly

Speaking in a loud and high-pitched tone

paralinguistic

[sigh]

Sighing sound

[inhale]

Inhaling sound

[laugh]

Laughter sound

[chuckle]

Chuckling sound

[exhale]

Exhaling sound

[clears throat]

Throat clearing sound

[snort]

Snorting sound

[giggle]

Giggling sound

[cough]

Coughing sound

[breath]

Breathing sound

[uhm]

Hesitation sound: "Uhm"

[Confirmation-en]

Confirming: "En"

[Surprise-oh]

Expressing surprise: "Oh"

[Surprise-ah]

Expressing surprise: "Ah"

[Surprise-wa]

Expressing surprise: "Wa"

[Surprise-yo]

Expressing surprise: "Yo"

[Dissatisfaction-hnn]

Dissatisfied sound: "Hnn"

[Question-ei]

Questioning: "Ei"

[Question-ah]

Questioning: "Ah"

[Question-en]

Questioning: "En"

[Question-yi]

Questioning: "Yi"

[Question-oh]

Questioning: "Oh"

Feature Requests & Wishlist

💡 We welcome all ideas for new features! If you'd like to see a feature added to the project, please start a discussion in our Discussions section.

We'll be collecting community feedback here and will incorporate popular suggestions into our future development plans. Thank you for your contribution!

Demos

Task

Text

Source

Edited

Emotion-Fear

我总觉得，有人在跟着我，我能听到奇怪的脚步声。

fear_zh_female_prompt.webm

fear_zh_female_output.webm

Style-Whisper

比如在工作间隙，做一些简单的伸展运动，放松一下身体，这样，会让你更有精力。

whisper_prompt.webm

whisper_output.webm

Style-Act_coy

我今天想喝奶茶，可是不知道喝什么口味，你帮我选一下嘛，你选的都好喝～

act_coy_prompt.webm

act_coy_output.webm

Paralinguistics

你这次又忘记带钥匙了 [Dissatisfaction-hnn]，真是拿你没办法。

paralingustic_prompt.webm

paralingustic_output.webm

Denoising

Such legislation was clarified and extended from time to time thereafter. No, the man was not drunk, he wondered how we got tied up with this stranger. Suddenly, my reflexes had gone. It's healthier to cook without sugar.

denoising_prompt.webm

denoising_output.webm

Speed-Faster

上次你说鞋子有点磨脚，我给你买了一双软软的鞋垫。

speed_faster_prompt.webm

speed_faster_output.webm

For more examples, see demo page.

Model Download

Model Usage

📜 Requirements

The following table shows the requirements for running Step-Audio-EditX model (batch size = 1):

| Model | Parameters | Setting
(sample frequency) | GPU Optimal Memory | |------------|------------|--------------------------------|----------------| | Step-Audio-EditX | 3B| 41.6Hz | 12 GB |

• An NVIDIA GPU with CUDA support is required.
  • The model is tested on a single L40S GPU.
  • 12GB is just a critical value, and 16GB GPU memory shoule be safer.
• Tested operating system: Linux

🔧 Dependencies and Installation

• Python >= 3.12
• PyTorch >= 2.9.1
• CUDA Toolkit

git clone https://github.com/stepfun-ai/Step-Audio-EditX.git

cd Step-Audio-EditX
uv sync --refresh
source .venv/bin/activate

git lfs install
git clone https://huggingface.co/stepfun-ai/Step-Audio-Tokenizer
git clone https://huggingface.co/stepfun-ai/Step-Audio-EditX
git clone https://huggingface.co/stepfun-ai/Step-Audio-EditX-AWQ-4bit/

After downloading the models, where_you_download_dir should have the following structure:

where_you_download_dir
├── Step-Audio-Tokenizer
├── Step-Audio-EditX

Run with Docker

You can set up the environment required for running Step-Audio-EditX using the provided Dockerfile.

# build docker
docker build . -t step-audio-editx

# run docker
docker run --rm --gpus all \
    -v /your/code/path:/app \
    -v /your/model/path:/model \
    -p 7860:7860 \
    step-audio-editx

Local Inference Demo

[!TIP] For optimal performance, keep audio under 30 seconds per inference.

# zero-shot cloning
# The path of the generated audio file is output/fear_zh_female_prompt_cloned.wav
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "我总觉得，有人在跟着我，我能听到奇怪的脚步声。" \
    --prompt-audio "examples/fear_zh_female_prompt.wav" \
    --generated-text "可惜没有如果，已经发生的事情终究是发生了。" \
    --edit-type "clone" \
    --output-dir ./output 

python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "His political stance was conservative, and he was particularly close to margaret thatcher." \
    --prompt-audio "examples/zero_shot_en_prompt.wav" \
    --generated-text "Underneath the courtyard is a large underground exhibition room which connects the two buildings.	" \
    --edit-type "clone" \
    --output-dir ./output 

# edit
# There will be one or multiple wave files corresponding to each edit iteration, for example: output/fear_zh_female_prompt_edited_iter1.wav, output/fear_zh_female_prompt_edited_iter2.wav, ...
# emotion; fear
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "我总觉得，有人在跟着我，我能听到奇怪的脚步声。" \
    --prompt-audio "examples/fear_zh_female_prompt.wav" \
    --edit-type "emotion" \
    --edit-info "fear" \
    --output-dir ./output 

# emotion; happy
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "You know, I just finished that big project and feel so relieved. Everything seems easier and more colorful, what a wonderful feeling!" \
    --prompt-audio "examples/en_happy_prompt.wav" \
    --edit-type "emotion" \
    --edit-info "happy" \
    --output-dir ./output 

# style; whisper
# for style whisper, the edit iteration num should be set bigger than 1 to get better results.
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "比如在工作间隙，做一些简单的伸展运动，放松一下身体，这样，会让你更有精力." \
    --prompt-audio "examples/whisper_prompt.wav" \
    --edit-type "style" \
    --edit-info "whisper" \
    --output-dir ./output 

# paraliguistic 
# supported tags, Breathing, Laughter, Surprise-oh, Confirmation-en, Uhm, Surprise-ah, Surprise-wa, Sigh, Question-ei, Dissatisfaction-hnn
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "我觉得这个计划大概是可行的，不过还需要再仔细考虑一下。" \
    --prompt-audio "examples/paralingustic_prompt.wav" \
    --generated-text "我觉得这个计划大概是可行的，[Uhm]不过还需要再仔细考虑一下。" \
    --edit-type "paralinguistic" \
    --output-dir ./output 

# denoise
# Prompt text is not needed.
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-audio "examples/denoise_prompt.wav"\
    --edit-type "denoise" \
    --output-dir ./output 

# vad 
# Prompt text is not needed.
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-audio "examples/vad_prompt.wav" \
    --edit-type "vad" \
    --output-dir ./output 

# speed
# supported edit-info: faster, slower, more faster, more slower
python3 tts_infer.py \
    --model-path where_you_download_dir \
    --tokenizer-path where_you_download_dir \
    --prompt-text "上次你说鞋子有点磨脚，我给你买了一双软软的鞋垫。" \
    --prompt-audio "examples/speed_prompt.wav" \
    --edit-type "speed" \
    --edit-info "more faster" \
    --output-dir ./output 

Launch Web Demo

Start a local server for online inference. Assume you have one GPU with at least 12GB memory available and have already downloaded all the models.

# Standard launch
python app.py --model-path where_you_download_dir --tokenizer-path where_you_download_dir --model-source local

# Using pre-quantized AWQ 4-bit models, memory-efficient mode (for limited GPU memory, ~6-8GB usage)
python app.py \
    --model-path path/to/quantized/model \
    --tokenizer-path where_you_download_dir \
    --model-source local \
    --gpu-memory-utilization 0.1 \
    --enforce-eager \
    --max-num-seqs 1 \
    --cosyvoice-dtype bfloat16 \
    --no-cosyvoice-cuda-graph

Available Parameters

Memory Usage Guide

Training

Please refer to script/ReadMe.md

🔄 Model Quantization (Optional)

For users with limited GPU memory, you can create quantized versions of the model to reduce memory requirements:

# Create an AWQ 4-bit quantized model
python quantization/awq_quantize.py --model_path path/to/Step-Audio-EditX

# Advanced quantization options
python quantization/awq_quantize.py

For detailed quantization options and parameters, see quantization/README.md.

Technical Details

Step-Audio-EditX comprises three primary components:

• A dual-codebook audio tokenizer, which converts reference or input audio into discrete tokens.
• An audio LLM that generates dual-codebook token sequences.
• An audio decoder, which converts the dual-codebook token sequences predicted by the audio LLM back into audio waveforms using a flow matching approach.

Audio-Edit enables iterative control over emotion and speaking style across all voices, leveraging large-margin data during SFT and PPO training.

Evaluation

Comparison between Step-Audio-EditX and Closed-Source models.

• Step-Audio-EditX demonstrates superior performance over Minimax and Doubao in both zero-shot cloning and emotion control.
• Emotion editing of Step-Audio-EditX significantly improves the emotion-controlled audio outputs of all three models after just one iteration. With further iterations, their overall performance continues to improve.

Generalization on Closed-Source Models.

• For emotion and speaking style editing, the built-in voices of leading closed-source systems possess considerable in-context capabilities, allowing them to partially convey the emotions in the text. After a single editing round with Step-Audio-EditX, the emotion and style accuracy across all voice models exhibited significant improvement. Further enhancement was observed over the next two iterations, robustly demonstrating our model's strong generalization.

• For paralinguistic editing, after editing with Step-Audio-EditX, the performance of paralinguistic reproduction is comparable to that achieved by the built-in voices of closed-source models when synthesizing native paralinguistic content directly. (sub means replacement of paralinguistic tags with native words)

  Table: Generalization of Emotion, Speaking Style, and Paralinguistic Editing on Closed-Source Models.

Language

Model

Emotion ↑

Speaking Style ↑

Paralinguistic ↑

Iter0

Iter1

Iter2

Iter3

Iter0

Iter1

Iter2

Iter3

Iter0

sub

Iter1

Chinese

MiniMax-2.6-hd

71.6

78.6

81.2

83.4

36.7

58.8

63.1

67.3

1.73

2.80

2.90

Doubao-Seed-TTS-2.0

67.4

77.8

80.6

82.8

38.2

60.2

65.0

64.9

1.67

2.81

2.90

GPT-4o-mini-TTS

62.6

76.0

77.0

81.8

45.9

64.0

65.7

69.7

1.71

2.88

2.93

ElevenLabs-v2

60.4

74.6

77.4

79.2

43.8

63.3

69.7

70.8

1.70

2.71

2.92

English

MiniMax-2.6-hd

55.0

64.0

64.2

66.4

51.9

60.3

62.3

64.3

1.72

2.87

2.88

Doubao-Seed-TTS-2.0

53.8

65.8

65.8

66.2

47.0

62.0

62.7

62.3

1.72

2.75

2.92

GPT-4o-mini-TTS

56.8

61.4

64.8

65.2

52.3

62.3

62.4

63.4

1.90

2.90

2.88

ElevenLabs-v2

51.0

61.2

64.0

65.2

51.0

62.1

62.6

64.0

1.93

2.87

2.88

Average

MiniMax-2.6-hd

63.3

71.3

72.7

74.9

44.2

59.6

62.7

65.8

1.73

2.84

2.89

Doubao-Seed-TTS-2.0

60.6

71.8

73.2

74.5

42.6

61.1

63.9

63.6

1.70

2.78

2.91

GPT-4o-mini-TTS

59.7

68.7

70.9

73.5

49.1

63.2

64.1

66.6

1.81

2.89

2.90

ElevenLabs-v2

55.7

67.9

70.7

72.2

47.4

62.7

66.1

67.4

1.82

2.79

2.90

Table: Generalization of Emotion, Speaking Style, and Paralinguistic Editing on Step-Audio-EditX.

Language

Model

Emotion ↑

Speaking Style ↑

Paralinguistic ↑

Iter0

Iter1

Iter2

Iter3

Iter0

Iter1

Iter2

Iter3

Iter0

Iter1

Chinese

20251112

57.0

71.7

74.5

77.7

41.6

62.1

65.8

69.2

1.80

2.89

20251128

58.7

73.6

75.1

77.8

40.4

62.1

65.3

68.0

1.80

2.89

20260129

60.1

75.0

79.1

81.6

51.1

70.0

68.9

62.4

2.07

2.91

English

20251112

49.9

60.5

61.5

63.7

50.3

62.4

64.3

63.1

2.02

2.88

20251128

51.2

60.0

63.1

64.2

48.8

63.4

62.3

64.4

2.02

2.89

20260129

51.0

63.1

65.5

67.0

43.3

60.4

66.5

69.6

2.18

2.93

Average

20251112

53.5

66.1

68.0

70.7

46.0

62.3

65.1

66.2

1.91

2.89

20251128

55.0

66.8

69.1

71.0

44.6

62.8

63.8

66.2

1.91

2.89

20260129

55.6

69.1

72.3

74.3

47.2

65.2

67.7

66.0

2.12

2.92

Acknowledgements

Part of the code and data for this project comes from:

• CosyVoice
• transformers
• FunASR
• NVSpeech
• vllm

Thank you to all the open-source projects for their contributions to this project!

License Agreement

• The code in this open-source repository is licensed under the Apache 2.0 License.

Citation

@misc{yan2025stepaudioeditxtechnicalreport,
      title={Step-Audio-EditX Technical Report}, 
      author={Chao Yan and Boyong Wu and Peng Yang and Pengfei Tan and Guoqiang Hu and Yuxin Zhang and Xiangyu and Zhang and Fei Tian and Xuerui Yang and Xiangyu Zhang and Daxin Jiang and Gang Yu},
      year={2025},
      eprint={2511.03601},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2511.03601}, 
}

⚠️ Usage Disclaimer

• Do not use this model for any unauthorized activities, including but not limited to:
  • Voice cloning without permission
  • Identity impersonation
  • Fraud
  • Deepfakes or any other illegal purposes
• Ensure compliance with local laws and regulations, and adhere to ethical guidelines when using this model.
• The model developers are not responsible for any misuse or abuse of this technology.

We advocate for responsible generative AI research and urge the community to uphold safety and ethical standards in AI development and application. If you have any concerns regarding the use of this model, please feel free to contact us.

Star History