Microphone (ALSA)
    │
    ▼
Voice Activity Detection (Silero VAD)
    │  Buffers speech, detects silence timeout
    ▼
Speech-to-Text (Whisper Tiny EN)
    │
    ▼
Large Language Model (Qwen3 1.7B Q8)
    │
    ▼
Text-to-Speech (Piper Lessac Medium)
    │
    ▼
Speaker (ALSA)

linux-voice-assistant/
├── src/
│   ├── main.cpp                    # Entry point, CLI parsing, main loop
│   ├── audio/
│   │   ├── audio_capture.h/cpp     # ALSA mic input (16kHz, 16-bit PCM, mono)
│   │   └── audio_playback.h/cpp    # ALSA speaker output (multi-rate)
│   ├── pipeline/
│   │   └── voice_pipeline.h/cpp    # Full pipeline: VAD -> STT -> LLM -> TTS
│   └── config/
│       └── model_config.h          # Model paths, IDs, availability checks
├── tests/
│   └── test_pipeline.cpp           # Feed WAV file through pipeline (no mic needed)
├── scripts/
│   └── download-models.sh          # Download all required models
├── CMakeLists.txt                  # Build configuration
├── build.sh                        # End-to-end build script
└── README.md

# 1. Build everything (SDK + models + assistant)
./build.sh

# 2. Run the voice assistant
./build/voice-assistant

# Step 1: Download Sherpa-ONNX
cd ../../sdk/runanywhere-commons
./scripts/linux/download-sherpa-onnx.sh

# Step 2: Build runanywhere-commons
./scripts/build-linux.sh

# Step 3: Download models
cd ../../Playground/linux-voice-assistant
./scripts/download-models.sh

# Step 4: Build
mkdir -p build && cd build
cmake ..
cmake --build . -j$(nproc)

# Step 5: Run
./voice-assistant

# Required models (VAD, STT, LLM, TTS)
./scripts/download-models.sh

# Download an additional LLM bundle:
./scripts/download-models.sh --llm llama-3.2-3b
./scripts/download-models.sh --llm qwen3-4b

# Basic usage
./build/voice-assistant

# Select audio devices
./build/voice-assistant --list-devices
./build/voice-assistant --input hw:1,0 --output hw:0,0

# Test pipeline with a WAV file (no microphone needed)
./build/test-pipeline path/to/audio.wav