Orpheus TTS is a family of open-weight text-to-speech models from Canopy Labs that aims at human-sounding speech: built on an LLM backbone, the models deliver expressive, low-latency voice generation that can be steered with emotion cues and non-verbal sounds. Orpheus has been downloaded over a million times, forked and fine-tuned thousands of times, and adopted by large companies, according to Canopy Labs. The GitHub repository has about 6,400 stars, and a multilingual model family followed the original English release in a 2025 research preview.
Most open TTS systems sound flat because they treat speech as a pure signal problem. Orpheus instead fine-tunes a language model on speech tokens, so it understands context and can produce natural intonation, emphasis and paralinguistic details such as laughter or sighs when prompted with tags. That design makes it suitable for real-time, interactive applications: the model streams audio quickly enough for conversational agents and live avatars, which is the direction Canopy Labs is pursuing with its realtime avatar models. Optimized inference paths, including an fp8 deployment with Baseten, make production serving practical, and a published training guide explains how the multilingual versions were created so others can extend Orpheus to new languages.
Orpheus TTS is free and open source: the models and code are released under the Apache-2.0 license, so you can download the weights and run them on your own hardware at no cost beyond compute.
Orpheus TTS is for developers building voice agents, interactive avatars, audiobook and content pipelines, and researchers who want an expressive open-weight speech model they can fine-tune for new voices and languages.