Voices
Tara, Leah, Jess, Leo, Dan, Mia, Zac and Zoe
Language
English
Tags
Laugh, chuckle, sigh, cough, sniffle, groan, yawn, gasp
Voice training
Yes, from 1 to 30 minutes of speech
Memory
About 5.5 GB
Voice cloning
Not available; train a voice instead
Made by
Canopy Labs

Orpheus from Canopy Labs is tuned for conversation: it sounds like someone talking rather than reading.

Voices

Eight presets, in Canopy Labs’ order of realism: Tara (the default), Leah, Jess, Leo, Dan, Mia, Zac and Zoe.

Tags

Add a laugh, chuckle, sigh, cough, sniffle, groan, yawn or gasp exactly where you want it, from the script’s tag button.

Train

Orpheus doesn’t clone from a clip, but it can learn. Give it one to thirty minutes of your speech in 5 to 900 clips and Avocado trains a voice on your Mac. The eight presets stay available alongside it.

Long scripts

Orpheus speaks long scripts in sentence-sized pieces that share one voice, so a long read is never cut short.

Licence

Orpheus’s licence is set by Canopy Labs. Read it on the Orpheus model page before you publish anything you make.

Questions

Can Orpheus speak in my voice?

Not from a short clip; Orpheus doesn't clone. Instead, train it on one to thirty minutes of your own recordings and it learns your voice.

Does the model card require anything?

Canopy Labs asks you to accept its terms on the model page before you can get the model. Avocado works with it once it's on your Mac.

Avocado is almost here

Free for Apple silicon Macs running macOS 15 or later. Coming soon.

Coming soon for Mac