Step-Audio-EditX is a revolutionary AI audio editing model that makes "editing voice with natural language instructions" a seamless experience. This tool makes voice editing as intuitive and efficient as revising a text document. Users can precisely control the emotion, style, and intricate effects of their audio simply by typing simple text commands into Step-Audio-EditX.
Generate NowExperience text-driven audio editing and instantly generate your desired effects.
For example, if a user has a monotonous narration and wants to make it livelier, with Step-Audio-EditX, they just add an emotion tag (like adding [Happy]) to the text instruction. The model will adjust the voice to sound joyous. Adding paralinguistic effects like laughter will make the voice sound even more realistic. Users can even iterate multiple times, gradually enhancing the effect until the final audio meets their exact expectations. This is the power of Step-Audio-EditX.
Source Audio:
“我总觉得，有人在跟着我，我能听到奇怪的脚步声。”
user prompt：
Fear Emotion
Edit Output:
“我总觉得，有人在跟着我，我能听到奇怪的脚步声。” (Fear)
Source Audio:
“你到底想怎么着，上学的时候懒得学，工作的时候没时间学。”
user prompt：
Roar Speaking Style
Edit Output:
“你到底想怎么着，上学的时候懒得学，工作的时候没时间学。”
Source Audio:
“Wait, you're telling me you finished the entire book in one day? That's incredible!”
user prompt：
Add breathing sounds paralinguistic features
Edit Output:
Wait, you're telling me you finished the entire book in one day? [Surprise-oh] That's incredible!
Source Audio:
“就是说你比如说我一共在这次看病我一共花了一百块钱，其中呢医生的这个劳动价值占了三十块钱。 ”
user prompt：
Silence Trimming
Edit Output:
“就是说你比如说我一共在这次看病我一共花了一百块钱，其中呢医生的这个劳动价值占了三十块钱。”
Step-Audio-EditX boasts a rich set of audio editing capabilities, meeting various voice processing needs:
Instantly imitate any speaker's voice to read text without extra training; supports Mandarin, English, plus dialects like Sichuanese and Cantonese, enabling cross-lingual audio creation.
Allows precise adjustment of voice emotion (e.g., anger, joy) and speaking style (e.g., serious, whisper) for the same sentence to deliver different expressive effects.
Step-Audio-EditX can insert subtle details like breathing, laughter, and sighs, making the synthesized voice closer to human conversation and more expressive.
Supports multi-step continuous editing, allowing users to gradually enhance the target emotion or effect, maintaining meticulous control over the final audio output.
The functions of Step-Audio-EditX are applicable across a wide range of scenarios, making it highly versatile:
Short video creators can switch characters' voices (e.g., to a lively girl) with one click; audiobook authors can dub multi-character dialogues solo, making works more engaging and dynamic.
With Step-Audio-EditX, virtual assistants can dynamically adjust tone (e.g., friendly, concerned) and add subtle details like breathing, making the voice sound natural and lively—no more robotic responses!
Film and game voice actors can use Step-Audio-EditX for zero-shot cloning of their voices and subtle emotion adjustments to match scenes; dialogue can also be instantly converted to different languages or dialects, dramatically cutting localization costs.
Step-Audio-EditX is based on a large language model, achieving unified cross-modal editing. It's lightweight and efficient, with a compact 3 billion parameters, capable of running on a single GPU, making it easy to integrate into various applications.
Core Parameter Count
3 Billion
Training Data Type
Multi-lingual Emotional Voice
Key Metric Improvement
MOS Score 4.3 / BLEU Increase 17%
Get started with Step-Audio-EditX now.