Orpheus TTS canopyai/Orpheus-TTS

Llama-based 3B speech LLM from Canopy Labs for expressive, emotive, low-latency streaming text-to-speech.

Creator kit + Compare GitHub ↗ Install guide (official) ↗

Orpheus TTS at a glance

Checked 15 Oct 2026

Orpheus TTS is a 3B-parameter speech LLM built on Llama-3.2-3B-Instruct, with English finetuned voices, emotive tags like <laugh> and <sigh>, about 200ms streaming latency, and a multilingual research preview. Code is Apache-2.0; weights are tagged Apache-2.0 on Hugging Face but are gated and list a Llama base model. Main inference uses vLLM; a llama.cpp path runs without a GPU. Zero-shot cloning is claimed but the cloning notebook is still on the to-do list.

Cite this pageRepoQuarry. “Orpheus TTS.” Checked 15 Oct 2026. https://repoquarry.dev/projects/orpheus-tts

How much of ElevenLabs does it replace?

2 of 4 jobs. Expressive streaming speech with preset voices and emotion tags. Zero-shot cloning is claimed, but the README lists fixing the voice cloning notebook as unfinished.

Speech generation Partial replacement

Voice cloning Partial replacement

Speech recognition Not covered

Voice agents Not covered

For speech recognition, voice agents, see other ElevenLabs alternatives. You may also need to build an interface, an API server and hosting yourself.

Facts, with sources

Every answer links to where we found it. We read the documentation; we did not install it.

Report a correction

QuestionAnswerSourceChecked

Runs on my machine? Yes, GPU via vLLM, or CPU via the llama.cpp-based orpheus-cpp You can stream audio without a GPU by using orpheus-cpp, which is a llama.cpp-compatible backend of the Orpheus TTS model. github.com · README.md ↗ 1d ago

Code licence Apache-2.0 github.com · LICENSE ↗ 1d ago

Model weights licence Tagged Apache-2.0 on Hugging Face, but weights are gated and list meta-llama/Llama-3.2-3B-Instruct as a base model, whose own licence may also apply Orpheus TTS is a state-of-the-art, Llama-based Speech-LLM designed for high-quality, empathetic text-to-speech generation. huggingface.co · orpheus-3b-0.1-ft ↗ 1d ago

Commercial use Likely yes under Apache-2.0, but check the Llama 3.2 base-model terms and the model card's misuse rules Do not use our models for impersonation without consent, misinformation or deception (including fake news or fraudulent calls), or any illegal or harmful activity. huggingface.co · orpheus-3b-0.1-ft ↗ 1d ago

Paid services needed? None required; Baseten hosted deployment is offered as an optional extra Baseten is our preferred inference partner for Orpheus. github.com · Orpheus-TTS#one-click-deployment-on-baseten ↗ 1d ago

Hardware GPU path via vLLM; a no-GPU path via llama.cpp is documented; no VRAM or RAM numbers stated github.com · README.md ↗ 1d ago

Operating systems CPU path documents Linux, Windows and Apple Silicon macOS; OS support for the main vLLM path is not stated github.com · README.md ↗ 1d ago

Setup effort Medium, pip install orpheus-speech (uses vLLM, may need vllm==0.7.3) plus Hugging Face access to the gated weights vllm pushed a slightly buggy version on March 18th so some bugs are being resolved by reverting to pip install vllm==0.7.3 after pip install orpheus-speech github.com · Orpheus-TTS#streaming-inference-example ↗ 1d ago