AXIOM - Advanced Voice Agent with Conversational Intelligence

AXIOM is a powerful and production-grade voice-first AI system designed specifically for robotics labs, operating fully offline without the need for API keys. It leverages real-time speech processing, intelligent intent classification, and context-aware responses to deliver an exceptional user experience with latency under 400ms. Optimized for systems with 4GB VRAM, especially the GTX 1650, it excels in environments demanding low-latency voice interaction.

Key Features

• Instant Voice Interaction: Interactive voice command processing utilizing WebSocket communication for seamless exchanges.
• Intelligent Intent Classification: Achieves over 88% confidence using a SetFit-based model to accurately detect user intents.
• Context-Aware Responses: Employs Semantic RAG with over 2,116 template responses, enabling accurate reply generation based on conversational context.
• 3D Interactive UI: Integrates a WebGL-based carousel for intuitive visual interactions with robotic equipment.
• Multi-Turn Conversation: Utilizes FIFO history management to maintain context across interactions, enhancing conversational coherence.
• Optimized Latency: Ensures sub-2 second response times for real-time, interactive experiences.
• High-Quality Text-to-Speech: Incorporates phonetic processing and minimal corrections for natural-sounding speech output.
• Future-Ready Training: Logs interaction data to facilitate constant improvement.

Four Breakthrough Features

1. Glued Interactions: Maintains a FIFO history of the last five interactions, injecting previous context into the conversation for continuity.
2. Zero-Copy Inference: Direct tensor streaming significantly enhances memory efficiency, reducing overall latency by 2.4%.
3. 3D Holographic UI: Implements an innovative WebGL-based visualization that dynamically loads 3D models based on user interactions.
4. Dual Corrector Pipeline: Provides clean and articulate TTS output through phonetic and minimal safe corrections.

Performance Metrics

Comprehensive benchmarks demonstrate AXIOM’s high performance across various metrics, including substantial reductions in latency and memory usage compared to traditional methods.

System Architecture

The AXIOM architecture comprises a FastAPI backend, handling components like Speech-to-Text (STT), intent classification, and Text-to-Speech (TTS), all enabling efficient real-time processing in a structured manner.

How It Works

AXIOM captures voice input through a web interface, processes it via a sophisticated inference pipeline, and generates responses using a combination of template and responsive generative AI, ensuring quick and accurate output.

Live Demos and Screenshots

Visual demos showcase the web interface and demonstrate the seamless real-time interaction capabilities.

Community and Contribution

AXIOM encourages contributions, with clear guidelines for participation, an active issue tracking for enhancements, and a code of conduct to foster a collaborative environment.

Incorporate AXIOM into robotics labs for a robust, intelligent voice agent capable of powering interactive experiences and enhancing productivity.