71 lines
2.3 KiB
Plaintext
71 lines
2.3 KiB
Plaintext
# Core
|
||
pyyaml>=6.0
|
||
ruamel.yaml>=0.18 # comment-preserving config.yaml writes from the GUI
|
||
loguru>=0.7
|
||
httpx>=0.27
|
||
numpy>=1.24,<2.0
|
||
pydantic>=2.5
|
||
|
||
# Audio I/O
|
||
sounddevice>=0.4.6
|
||
soundfile>=0.12
|
||
scipy>=1.11
|
||
|
||
# Wake word + VAD
|
||
silero-vad>=5.1
|
||
openwakeword>=0.6.0
|
||
onnxruntime>=1.17
|
||
|
||
# STT
|
||
# BatchedInferencePipeline (4× GPU speedup) requires >=1.1; Silero-VAD v6 in >=1.2.1
|
||
faster-whisper>=1.1
|
||
|
||
# HuggingFace – hf_xet enables fast Xet-storage downloads (avoids HTTP fallback warning)
|
||
hf_xet>=1.0
|
||
|
||
# Speaker identification
|
||
speechbrain>=1.0
|
||
|
||
# TTS — Piper VITS (CPU ONNX inference, fast and low-overhead)
|
||
# piper-tts 2.x requires Python >=3.11; 1.7.0 has the same API surface
|
||
piper-tts>=1.7.0,<2.0
|
||
# For voice training only (not needed for inference):
|
||
# pip install piper-train
|
||
# pip install onnx "numpy<2" (ONNX export; keep numpy<2 — newer onnx deps pull numpy 2)
|
||
# espeak-ng is required for training (German phonemisation):
|
||
# https://github.com/espeak-ng/espeak-ng/releases — install the Windows binary and add to PATH
|
||
|
||
# MCP
|
||
mcp>=1.2,<3
|
||
|
||
# Slave-device API server (javis-slave companion app)
|
||
fastapi>=0.110
|
||
uvicorn>=0.29
|
||
python-multipart>=0.0.9
|
||
|
||
# Optional: F5-TTS voice cloning backend (pip install f5-tts)
|
||
# Requires CC-BY-NC model weights downloaded from Hugging Face on first run.
|
||
# Switch backend via: tts.backend: f5 in config.yaml
|
||
# f5-tts>=1.1.20
|
||
|
||
# Optional: DeepFilterNet neural noise reduction (pip install deepfilternet)
|
||
# Automatically denoises mic input before STT when installed.
|
||
# deepfilternet>=0.5.6
|
||
|
||
# GUI
|
||
PySide6>=6.6
|
||
qasync>=0.27
|
||
|
||
# CUDA (install separately based on driver version)
|
||
# CUDA 13.2 (current): pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu132
|
||
# CUDA 12.4: pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu124
|
||
# CUDA 12.1: pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu121
|
||
# Or let the app auto-detect: python -m javis --install-deps
|
||
#
|
||
# NOTE: CTranslate2 (faster-whisper STT) requires cublas64_12.dll at runtime.
|
||
# Option A (pip — lightweight, recommended):
|
||
# pip install nvidia-cublas-cu12 nvidia-cudnn-cu12 nvidia-cuda-runtime-cu12
|
||
# Option B: install the full CUDA 12 Toolkit from NVIDIA:
|
||
# https://developer.nvidia.com/cuda-12-0-1-download-archive
|
||
# Javis auto-discovers option-A DLLs via os.add_dll_directory() at startup.
|