autonet-voice-core 0.1.0

Shared low-level audio primitives for the kevin and autonet voice services

voice-core

Shared low-level audio primitives for the kevin voice service and the autonet (autonet-computer[voice]) VoiceService. Both repos used to carry their own copy of this code; this package is the single source of truth so fixes land once.

What's here

Module	Contents
voice_core.audio	AudioMixer, AudioChannel (real-time mixing, ducking, fade, per-channel pause), find_device, get_device_list, resample
voice_core.tts	generate_kokoro / generate_edge / generate_elevenlabs / generate_piper; set_kokoro_model_dir, set_piper_module_dir
voice_core.stt	PushToTalkRecorder (duck/mute modes, 30s release-timeout safeguard), transcribe (faster-whisper)
voice_core.tones	make_tool_tone, make_result_chime, make_startup_chime, make_failure_tone, gen_tone
voice_core.text	strip_markdown (markdown + emoji stripping for TTS), tool_summary, tool_narration

What's not here

Architecture-specific glue stays in each repo:

• kevin: UDP instance registry, hooks, window focus, clipboard paste, overlay, main()
• autonet: VoiceService class, EventBus wiring, VoiceConfig

Install

pip install -e .            # core only (numpy + sounddevice)
pip install -e .[full]      # all TTS backends + PTT/STT

Heavy deps (kokoro-onnx, edge-tts, miniaudio, requests, keyboard, faster-whisper) are lazy-imported, so the core imports cheaply and each backend only errors (with an install hint) when used without its dependency.

Configuring model/voice paths

The Kokoro model files (kokoro-v1.0.onnx, voices-v1.0.bin) and the Piper voice package live outside this repo. Point at them at startup:

import voice_core
voice_core.set_kokoro_model_dir(r"c:\code\kevin")   # holds the .onnx + .bin
voice_core.set_piper_module_dir(r"c:\code\voice")   # holds the `voice` package