hey-clever

Record spoken commands, transcribe with Whisper, and reply via TTS.

3|Updated Jul 17, 2024

Search Tags:#whisper #tts #systemd #wake-word #speech-recognition #voice-assistant

Ready to run

npx skills add https://github.com/Castrozan/.dotfiles --skill hey-clever

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: hey-clever
Source: https://github.com/Castrozan/.dotfiles/tree/main/agents/skills/hey-clever
Command: npx skills add https://github.com/Castrozan/.dotfiles --skill hey-clever

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires numpy, onnxruntime, requests, sounddevice, openwakeword, faster_whisper, edge-tts, and includes scripts (resource) components.

What problem does it solve?

This Skill provides an always-on wake-word based voice assistant that records spoken commands, transcribes them with Whisper, and replies via TTS.

Core Features & Use Cases

Activation by wake word, automatic recording, and transcription using Whisper.
Local processing with Silero VAD for robust voice activity detection.
Transcription with Whisper tiny for keyword detection and Whisper small for full command understanding.
Gateway communication to a Clawdbot-like backend and TTS playback of responses.
Optional systemd user service for auto-start and seamless background operation.

Quick Start

Install dependencies via the included setup script to prepare a Python venv and user service.
Verify input devices with: python3 hey-clever.py --list-devices
Start the service with: systemctl --user start hey-clever

Frequently Asked Questions about hey-clever

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I build an always-on wake-word voice assistant using Whisper and TTS?▼

You can build an always-on wake-word voice assistant by using openwakeword for activation, recording commands with sounddevice, transcribing with faster_whisper, and replying via edge-tts. This skill handles the entire pipeline locally.

What do I need to run a local voice assistant with systemd auto-start?▼

To run a local voice assistant with systemd auto-start, you need Python 3, a virtual environment, and dependencies like numpy, onnxruntime, and sounddevice. The setup script configures the systemd user service for background operation.

How does wake-word detection work with faster_whisper for voice commands?▼

Wake-word detection works by using openwakeword to trigger recording, then transcribing audio with faster_whisper. The Whisper tiny model handles keyword detection, while Whisper small processes full command understanding for accurate responses.

Can I use openwakeword and edge-tts for local speech recognition and spoken replies?▼

Yes, you can use openwakeword for local wake-word detection and edge-tts for spoken replies. The skill integrates both libraries to capture spoken commands locally and output synthesized speech responses.

How do I list audio input devices for a Python voice assistant script?▼

To list audio input devices for a Python voice assistant, run the command python3 hey-clever.py --list-devices. This verifies available microphones before starting the sounddevice audio capture pipeline.

Does this voice assistant process audio completely offline?▼

The voice assistant processes wake-word detection and transcription locally using openwakeword and faster_whisper, but it communicates with a gateway backend and uses edge-tts, which may require network connectivity for responses.