Skip to main content
Rase AI Intelligence Newsletter

Creative AI Model or AI service

CosyVoice 2 brings streaming synthesis to text-to-speech research

CosyVoice 2 is a speech synthesis model designed to begin returning audio as text is processed. The team released a downloadable 500-million-parameter checkpoint. [2][3]

Model access
Downloadable CosyVoice2-0.5B model checkpoint. [3]
License
Apache License 2.0 as listed for the CosyVoice2-0.5B weights. [3]

What the model does

A December 2024 preprint describes CosyVoice 2’s streaming and non-streaming modes. The project lists the 0.5B release in December and links the weights. [1][2][3]

Why it matters

Streaming can shorten the wait for the first words in a read-aloud interface. A separate full-text workflow may be easier to review when an application has the complete passage ahead of time. [2]

Who can use it

Developers and speech researchers can evaluate the 0.5B checkpoint for reading interfaces or audio prototypes that need synthesized speech. [2][3]

Key dates

  1. CosyVoice 2 paper first submitted [2]

Ways to use it

Compare whether the model reads a long sentence naturally when given text in pieces and when given the full passage. Measure time to first audio alongside intelligibility. [2]

Model limitations

Audio timing and quality depend on input text and the serving environment. Check the exact checkpoint, because the model repository also describes newer CosyVoice releases. [1][2][3]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source Official CosyVoice project version and artifact history
  2. Primary source First version and submission history of the CosyVoice 2 report
  3. Primary source FunAudioLLM/CosyVoice2-0.5B
  4. Other source CosyVoice 2: Scalable Streaming Speech Synthesis with Large Language Models

Related profiles

Creative AI

Seedance 2.5 (4K API and Flash)

ByteDance (Volcano Engine / BytePlus)

These tools generate video through developer services. The cited sources describe different variants and routes of access.

Medium confidenceEvidence checked

Creative AI

LongCat-Video-Avatar 1.6

Meituan (LongCat team)

It generates video of an avatar from supported inputs in the documented project.

High confidenceEvidence checked

Creative AI

MiniMax H3-Image

MiniMax

It generates images from prompts through the model architecture described by MiniMax.

High confidenceEvidence checked