- Delivery
- Alibaba Cloud Model Studio API
- Integration
- DashScope SDK / WebSocket
- Voice sources
- System voices, cloning, voice design
- GitHub scope
- Client SDK; not downloadable model weights
Strengths worth exploring
Describe the performance
Use instruction to describe a calm documentary narrator or a lively character speaking a little faster. Direction can be applied to system voices and cloned voices.
Official source [1]Place acting cues in the script
Emotion tags such as [excited] and [sad] steer the following speech; [laughing] and [sighing] insert vocal effects at a chosen point. This is useful when one passage changes mood.
Official source [1]Direct dialects in cloned speech
The official guide shows instructions for Chinese dialect delivery with cloned voices. System voices instead need an appropriate language or dialect-capable voice selection.
Official source [1]Separate voice enrollment from delivery direction
Enrollment turns a reference recording into a Voice ID associated with a target model. Synthesis uses that ID after approval; changing the target model is not a simple voice-ID substitution.
language_hints helps identify language, while instruction describes delivery. Voice design creates a new voice from a description and is a different path from reference-based cloning.
Official source [2]A cloud product, with a client SDK
The GitHub link below is the official DashScope client SDK. It does not provide Qwen-Audio-3.0-TTS model weights. The separate Qwen3-TTS open-source project should not be presented as this cloud model's repository.
Official source [3]Usage and deployment
- On DS Speech, choose a public voice or create a private voice, enter a script, and set the delivery you want.
- For your own integration, enable the model in Alibaba Cloud Model Studio, obtain an API key, and select the correct region and workspace endpoints.
- The command below installs the client SDK only. Follow the official synthesis example with a model, voice, and text; complete enrollment and approval first for a cloned voice.
python -m venv .venv
source .venv/bin/activate
pip install dashscopeBefore you get started
- Features vary across model variants, voices, and regions. A family name alone does not establish which controls a specific voice supports.
- Check the documented streaming-mode restrictions for emotion and vocal-effect tags. Model capabilities do not imply every API mode is exposed by DS Speech.
Where it fits
Useful for creators directing Chinese dialects, expressive characters, and multilingual voiceovers through a managed cloud workflow.
Official resources and sources
DS Speech currently provides Qwen-Audio-3.0-TTS online. Other models in this library are documented for learning and self-deployment.