Echolet is a lightweight, offline voice assistant built for privacy and performance. It uses Whisper.cpp for voice recognition, Piper for ultra-fast text-to-speech, and a local LLM (via Ollama) for generating responses all running fully on your CPU without internet access.

🎯 Ideal for Linux users who want a private, fast, and customizable voice interaction system without depending on the cloud.

✅ Phase 1 Complete: Echolet currently supports full offline voice interaction including:

• Voice recording (mic input)
• Transcription using whisper.cpp
• LLM response via local models with Ollama
• Text-to-speech output using Piper

⚙️ Upcoming (Phase 2 and beyond):

• Voice-controlled commands (open apps, tell time)
• Notes and reminders (voice-to-memory)
• Wake word support
• File and document search
• Streamlit GUI

The project is fully modular and open for community contributions. Stay tuned for updates as Echolet evolves into a more complete offline personal assistant.

• Python 3.9+
• piper installed and in your PATH
• Ollama (for local LLM)
• Whisper.cpp (for speech-to-text)
• aplay (for audio playback)

Install Python dependencies:

pip install -r requirements.txt

1. Download a Piper Voice Model

• Choose a model from HuggingFace Piper Voices (e.g., en_US-amy-medium).

# Download model and config JSON
wget -O en_US-amy-medium.onnx "https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/amy/medium/en_US-amy-medium.onnx?download=true"
wget -O en_US-amy-medium.onnx.json "https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/amy/medium/en_US-amy-medium.onnx.json?download=true"

• Place them in:

/core/tts/piper/

2. Set PIPER_MODEL_PATH in config/config.py

PIPER_MODEL_PATH = "/absolute/path/to/en_US-amy-medium.onnx"

3. Install and Run Ollama

curl -fsSL https://ollama.com/install.sh | sh
ollama run phi

You can use ollama list to see installed models.

Start the current assistant:

python -m scripts.voice_assistant

Speak your prompt. Echolet will:

1. Listen using Whisper
2. Respond using a local LLM
3. Speak back using Piper

echolet/
├── main.py                 # Entry point
├── voice/
│   ├── record.py           # Audio recording logic
│   └── transcribe.py       # Whisper.cpp wrapper
├── llm/
│   └── query_model.py      # Ask questions using Ollama
├── tts/
│   └── speak.py            # TTS using Piper or pyttsx3
├── config/
│   └── config.py           # Paths and settings
└── requirements.txt        # Python dependencies

• If Piper fails, it falls back to pyttsx3 (no model needed).
• All modules are designed to run on low-end hardware without GPU.
• Ollama models like phi are ideal for 4GB+ RAM CPUs.

• Piper TTS
• Whisper.cpp
• Ollama
• pyttsx3

MIT License

Muhammad Saad