Sesame
Add emotionally intelligent, contextual voice to AI assistants.
Pricing Features
Free Preview
Open Source
Apache 2.0
Integration Features
GitHub Release
LLama Architecture Backbone
Mimi Split-RVQ Tokenizer
Capability Features
Text and Audio Input
TextAudio
Conversational Speech Generation
Emotional Intelligence
Conversational Dynamics
Context Awareness
Consistent Personality
Single-Stage Model
Evaluation Suite
Model Sizes
Tiny: 1B backbone, 100M decoderSmall: 3B backbone, 250M decoderMedium: 8B backbone, 300M decoder
Objective Metrics
Word Error RateSpeaker SimilarityHomograph DisambiguationPronunciation Consistency
Subjective Metrics
Comparative Mean Opinion Score
Multiple Speaker Handling
Pronunciation Correction
Dataset Size
1 million hours
Training Epochs
5
Sequence Length
2048
Partial Multilingual Support Planned
Planned for 20+ languages
Limitation Features
English Language Dominance
No Pre-trained Language Model Use
Cannot Model Conversation Structure
Real-Time Generation Delay
RVQ time-to-first-audio scales poorly
Memory Bottleneck in Training