Voice · Canopy Labs (open source)
Orpheus TTS
An open-source English voice model that can laugh, sigh or gasp on cue, running in private mode. Pick it for lively English dialogue.
Private High costWhat sets it apart
Orpheus is an Apache 2.0 speech model built on a Llama language model, with emotion tags such as laugh and sigh in the text and about 200 ms streaming latency. In Sub Rosa it speaks English only.
- Built on Llama 3.2 3B
- Emotion tags in the text
- About 200 ms streaming latency
Strengths and limits
Where it shines
- Tags like <laugh> and <sigh> in the text
- Private mode: nothing is kept
Think twice when
- English voices only
- 8 voices
- 8 voices, WAV output
- Up to 5,000 characters per request in Sub Rosa
Use cases
Lively English dialogue
Tara: <laugh> You really did that? Leo: <sigh> Yes, and I regret it.
Tags make laughs and sighs natural.
Private English narration
Read this short story aloud with feeling.
Private mode and expressive delivery.
Against its rivals
- Chatterbox HDChatterbox is also private and expressive, at a lower price.#7 of 10 on Artificial Analysis Speech Arena (1,102 Elo)
- KokoroKokoro is private and far cheaper, with 8 languages but less drama.#10 of 10 on Artificial Analysis Speech Arena (1,064 Elo)
- Qwen3 TTSQwen3 TTS is private too and covers 10 languages.
Version history
Each version, dated, with what it changed over the one before.
- Orpheus 3B 0.1 finetunedLatest
- Introduced open speech model with zero-shot cloning and emotion tags
The model in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Orpheus TTStts-orpheus | Voice | Private | $0.0281 per 1,000 characters$0.0843 a page read aloud |
Specifications
| Specification | Orpheus 3B |
|---|---|
| Released | March 2025 |
| Parameters | 3B |
| Weights | Open, Apache 2.0 |
| Accepts | text |
| Produces | audio |
| Languages | English in Sub Rosa |
| Price | $0.0281 per 1,000 characters |
| Privacy | Private |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.