Tango is an open-source family of diffusion models for text-to-audio generation from Declare Lab, turning written descriptions into sound effects, ambience and music. The repository has about 1,200 stars on GitHub and was updated in 2025. Note: this Tango is Declare Lab’s text-to-audio research project; it is a distinct project from other products named Tango that are listed on Tooloso.
Tango generates audio from text using a latent diffusion approach guided by a large language model: an instruction-tuned LLM interprets the prompt, and the diffusion model renders matching audio, which gives it strong prompt adherence for complex descriptions. The follow-up release, Tango 2, adds preference-based alignment (DPO) on the Audio-alpaca style data to better match human judgments of audio quality and text relevance. The project publishes its models on Hugging Face together with demos and the accompanying research papers, and it helped establish the modern text-to-audio field alongside models like AudioLDM. A newer sibling, TangoFlux, continues the line with much faster generation.
Tango was among the first open models to show that guidance from an instruction-tuned language model materially improves text-to-audio quality and controllability compared with plain text encoders. The inference code generates audio clips from a single prompt and can be fine-tuned on custom sound datasets, and the family’s checkpoints have been widely reused as baselines in subsequent text-to-audio research.
Tango is a free, open research release: the code and model weights are published on GitHub and Hugging Face at no cost for research and experimentation.
Tango is for audio researchers, sound designers and developers who want an open text-to-audio model they can run, study and fine-tune locally instead of relying on a closed API.