CosyVoice speech synthesis overview-Intelligent Speech Interaction(ISI)-阿里云帮助中心

CosyVoice combines text understanding and speech generation based on a large-scale pre-trained language model to convert text into natural speech. The service provides long-text speech synthesis, streaming text-to-speech, and voice cloning for applications such as intelligent customer service, audiobooks, in-car navigation, and educational tutoring.

Capabilities

Natural speech: CosyVoice uses a generative neural network speech model developed by Alibaba's Qwen Voice Lab to produce speech with natural intonation, rhythm, and emotional expression.
Diverse voices: Voices vary in gender, age, dialect, and vocal characteristics to suit applications such as news broadcasts and storytelling.
Real-time synthesis: The service synthesizes both long and short text in real time and streams the resulting audio.
Expressive speech: The service supports vocal events such as laughter and filler words, as well as speech with a range of emotions.

Features

Choose a feature based on how text is supplied and whether a custom voice is required.

Feature	Description	Integration documentation
Long-text speech synthesis	Accepts the complete text in a single input and returns the synthesized speech as a stream of binary audio data. Text must be UTF-8 encoded.	Long-text speech synthesis API reference
Streaming text-to-speech	Accepts text incrementally and returns audio in real time. For example, text generated by a large language model (LLM) can be sent as it becomes available, without waiting for the full text or concatenating it first.	Streaming text-to-speech API reference
Voice cloning	Extracts voice characteristics from a short audio clip to generate a similar, natural-sounding custom voice without a training process.	Voice cloning introduction and SDK code examples

When using long-text speech synthesis with a voice that supports Speech Synthesis Markup Language (SSML), you can use SSML to add background audio, insert pauses, and adjust pronunciation. For supported tags and usage, see Speech Synthesis Markup Language.

Billing and concurrency limits

CosyVoice speech synthesis is available only in the commercial version and does not offer a free trial. Activate the commercial version before using the service. For a comparison of the versions, see Trial and commercial versions. For activation instructions, see Upgrade to the commercial version.

For billing rules, see Billing methods.
For concurrency limits, see Concurrency and QPS.