A.R.E.S – Artificial Reasoning Engineered System

Ares is a fully local voice assistant that combines:

• 🎤 Whisper.cpp for speech-to-text (STT)
• 🧠 Llama.cpp for natural language processing (LLM)
• 🔊 Piper for text-to-speech (TTS)
• 👂 OpenWakeWord for wake word detection ("Hey Ares")

Everything runs offline — no internet is required for processing.
Ares is designed to be modular and extendable to control smart devices or even robots.

---

✨ Features (Current Progress)

✅ Wake Word ("Hey Ares/Ares")

• Powered by OpenWakeWord
• Starts listening only after hearing the wake phrase

✅ Speech-to-Text (STT) (CPU)

• Uses sounddevice + webrtcvad for smart recording (stops when you go quiet)
• Transcribes audio with whisper-cli (from Whisper.cpp)

✅ Local Language Model (LLM) (GPU)

• Runs llama.cpp in server mode
• Configurable system prompt → "Jarvis"-like personality

✅ Text-to-Speech (TTS)

• Piper HTTP server generates natural-sounding voices
• Multiple voices available (e.g., en_US-bryce-medium)

✅ Main Pipeline
Wake Word → Record → Transcribe → Send to LLM → Speak Response

✅ Benchmarking

• benchmark_ai.sh logs timings for STT, LLM, and TTS
• Results are stored in latency.md

✅ CI / Mock Mode

• GitHub Actions run Ares in Mock Mode (no audio hardware required)
• Simulates STT, LLM, and TTS responses for automated testing

---

Getting Started

1. Install dependencies

pip install -r requirements.txt

Also build:

• Whisper.cpp
• Llama.cpp
• Piper

2. Start LLM + TTS servers

./server/run_servers.sh

Say "Hey Ares", wait for the beep 🎵, then speak your command. Ares will listen, process locally, and respond with speech.

🧪 Development & Testing

Benchmark Latency

./docs/benchmark_ai.sh

📂 Project Structure

llmio/               # Input/output modules
 ├─ stt_whisper.py   # Speech-to-text
 ├─ tts_piper.py     # Text-to-speech
 ├─ llm_remote.py    # LLM client
 └─ wake_word.py     # Wake word listener

scripts/             # Helper scripts
 └─ run_servers.sh   # Start LLM and TTS servers

latency.md           # Benchmark results
ci_runner.py         # CI harness with mocks
main.py              # Main application loop

⚙️ Roadmap

Voice & Interaction

• Custom wake word — trainable per device/user.
• Custom voice — selectable TTS voice profiles.

Web & App Actions

• Open websites & apps on command

Devices & I/O

• Bluetooth device control — pair/connect/disconnect and volume controls.

Perception

• Visual detection — optional camera input for object/face/basic scene cues.

Architecture/Security

• Speaker recognition — per-user profiles for personalization/permissions.
• Remote access (multi-device clients)

Hardware

• Custom hardware build — mic array, LEDs, physical mute, action button.

UI/UX

• User interface for better user experience