Canopy Labs
State-of-the-art AI speech models that sound indistinguishable from real humans
Capability Features
Orpheus Speech Models
Medium (3B)Small (1B)Tiny (400M)Nano (150M)
Zero-Shot Voice Cloning
Guided Emotion and Intonation
Realtime Streaming
Input Streaming for Lower Latency
Open Source Release Planned
Llama Architecture
Llama
Pretrained and Finetuned Models
Pretrained modelsFinetuned models
Sample Finetuning Scripts
Text to Speech
Handles Disfluencies
Emotion Tags
normalslowcryingsleepysighchuckle
Training Data Volume
100k+ hours of speech, billions of text tokens
LLM-based Customizability
Streaming Inference Speed
Faster than playback on A100 40GB for 3B model
Model Tokenizer Type
Non-streaming (CNN-based) tokenizer
Sliding Window Detokenizer
Demo Availability
Integration Features
Python Package for Streaming
GitHub Repository Access
Hugging Face Model Access
Google Colab Notebook
Baseten 1-Click Deployment
LLama Ecosystem Support
Limitation Features
English Language Only
No Explicit Pricing Details
No API Mentioned
No Mention of File Format Support