EdgeSpeaker is an AI that runs locally on your device. The avatar listens, speaks, and records, while keeping data on your machine, instead of sending it to someone else's cloud.

Core Features:

• Real-time voice synthesis with HeadTTS and Kokoro
• Speech recognition with Whisper
• Local LLM processing with WebLLM and Llama 3.2
• 3D avatar rendering with ReadyPlayerMe and TalkingHead
• Local chat storage (conversations stay on your device)
• Data management (export/import conversations for backup or device transfer)

Controls:

• Space Key: Hold to speak (voice input)
• Enter: Send typed message
• Escape: Close modals or interrupt AI response
• Mouse: Left drag to rotate camera, right drag to pan, scroll to zoom

Built With:

• Three.js for 3D graphics and rendering
• WebGPU/WASM for AI model execution
• Transformers.js for browser-based ML inference
• ONNX Runtime for cross-platform AI processing

Last updated: 2025-06-25

Eligibility

Users must be 13 years or older to use this service.

Acceptable Use

By using EdgeSpeaker, you agree not to:

• Use automated systems (bots, scrapers) to access our service
• Attempt to overload our servers or infrastructure
• Hotlink to or redistribute our assets without permission
• Use the service for illegal, harmful, or malicious purposes
• Reverse engineer or attempt to extract our AI models

Service Availability

EdgeSpeaker is provided "as-is" without warranties. We may modify, suspend, or discontinue the service at any time without notice.

AI Content Disclaimer

AI-generated responses may contain inaccuracies. Users should verify important information independently.

Limitation of Liability

EdgeSpeaker is provided "as-is" without any warranties, express or implied. You use this service at your own risk. We shall not be liable for any damages, including but not limited to data loss, privacy breaches, or service interruptions.

Intellectual Property

EdgeSpeaker and its original content are protected by copyright and other laws. Third-party components are subject to their respective licenses.

Jurisdiction

This service may not be available in all jurisdictions. Use is subject to local laws and regulations.

Enforcement

We reserve the right to terminate access for violations of these terms.

Core AI Components:

• Meta Llama 3.2 (Custom License) - Large language model for conversation and reasoning
• WebLLM © MLC Team (Apache 2.0) - Browser-based large language model inference engine
• Kokoro TTS Model (Apache 2.0) - High-quality voice synthesis model
• Whisper (via Xenova) © OpenAI (MIT) - Speech recognition and transcription

Avatar & 3D Graphics:

• Ready Player Me (Commercial License) - 3D avatar creation and rendering system
• Three.js (MIT) - 3D graphics rendering and animation library
• TalkingHead (MIT) - 3D animated avatar with facial expressions and speech
• HeadTTS (MIT) - Text-to-speech synthesis with lip-sync support

Framework & Infrastructure:

• Transformers.js © Hugging Face, Inc. (Apache 2.0) - Machine learning model deployment for web browsers
• ONNX (Apache 2.0) - Machine learning model format and runtime
• ONNX Runtime © Microsoft (MIT) - Cross-platform ML model execution
• lil-gui (MIT) - Lightweight settings and configuration interface
• Mixamo by Adobe (Royalty-Free) - 3D character animations service

Edit the system prompt that defines the avatar's personality and behavior. Use {TODAY} as a placeholder for the current date.