Skip to content

3P.md

Latest commit

 

History

History
207 lines (157 loc) · 8.89 KB

3P.md

File metadata and controls

207 lines (157 loc) · 8.89 KB
Raw

AlienVox — Third-Party Notices & Open-Source Compliance

This document lists every external component used by AlienVox, grouped by function. It is maintained alongside python_app/requirements.txt and python_app/stacks.yaml.

AlienVox itself is released under the MIT License — see LICENSE.

1. OS Platform APIs (no distribution obligation)

These APIs are part of the Windows operating system and are accessed at runtime through standard COM/Win32 interfaces. No binary redistribution occurs; no attribution is required beyond this notice.

Component Description Provider Notes
Windows SAPI 5 Speech API — COM voice enumeration, speak, stop, pause Microsoft Built into Windows Vista+; accessed via win32com.client
Windows Speech Platform Server Runtime voices (higher-quality SAPI-compatible voices) Microsoft Optional runtime install; same COM interface as SAPI 5
Win32 Clipboard API OpenClipboard / GetClipboardData / SetClipboardData Microsoft Used for text capture fallback via pywin32
WM_COPY / AttachThreadInput Foreground-window text extraction without keyboard simulation Microsoft Tier 1 capture path in src/capture.py

2. ML Models (weights — not bundled, downloaded on demand)

Model weights are not included in this repository. They are downloaded automatically on first use from Hugging Face Hub or project release channels. Each model has its own open-source license for both the architecture code and the weights.

2.1 Kokoro-82M

Item Detail
Name Kokoro-82M
Hub ID hexgrad/Kokoro-82M (auto-fetched by KPipeline on first use)
Type Neural TTS — StyleTTS2-based, 82 M parameters
License (code) Apache 2.0
License (weights) Apache 2.0
Source https://huggingface.co/hexgrad/Kokoro-82M
Obligation Preserve Apache 2.0 NOTICE and copyright headers if redistributing weights.

2.2 Piper

Item Detail
Name Piper
Version piper-tts >= 1.2.0 (PyPI) + individual .onnx voice files
Type Fast offline neural TTS — ONNX Runtime inference, ~60 MB per voice
License (runtime) MIT
License (voice models) Per-voice — most are MIT or CC0. Check https://github.com/rhasspy/piper/blob/master/VOICES.md
Source https://github.com/rhasspy/piper
Obligation MIT — attribution in docs sufficient; verify per-voice license before redistribution.

2.3 Chatterbox 0.5B

Item Detail
Name Chatterbox TTS
Hub ID resemble-ai/chatterbox (auto-fetched by ChatterboxTTS.from_pretrained())
Type Expressive zero-shot neural TTS — 0.5 B parameters
License (code) MIT
License (weights) MIT
Source https://github.com/resemble-ai/chatterbox
Obligation MIT — attribution required if redistributing.

2.4 Dia 1.6B

Item Detail
Name Dia
Install Source — pip install git+https://github.com/nari-labs/dia.git
Type Dialogue-focused expressive TTS — 1.6 B parameters, GPU-oriented (~3.5 GB)
License (code) Apache 2.0
License (weights) Apache 2.0
Source https://github.com/nari-labs/dia
Obligation Preserve Apache 2.0 NOTICE file and copyright headers if redistributing.

2.5 F5-TTS

Item Detail
Name F5-TTS
Hub ID SWivid/F5-TTS (auto-fetched on first use, ~1.2 GB)
Type Flow-matching zero-shot TTS
License (code) MIT
License (weights) MIT
Source https://github.com/SWivid/F5-TTS
Obligation MIT — attribution required if redistributing.

2.6 OuteTTS 0.5B

Item Detail
Name OuteTTS
Hub ID OuteAI/OuteTTS-0.3-500M (auto-fetched on first use)
Type Lightweight local TTS — 0.5 B parameters
License (code) MIT
License (weights) MIT
Source https://github.com/edwko/OuteTTS
Obligation MIT — attribution required if redistributing.

2.7 VibeVoice-Realtime-0.5B (under evaluation — not yet integrated, see docs/issues/todo_006.md)

Item Detail
Name VibeVoice-Realtime-0.5B
Hub ID microsoft/VibeVoice-Realtime-0.5B
Install Not on PyPI — pip install "vibevoice[streamingtts] @ git+https://github.com/microsoft/VibeVoice.git"
Type Streaming TTS — Qwen2.5-0.5B backbone + σ-VAE tokenizer + diffusion decoder, single-speaker only, ~200–300ms to first audible speech
License (code) MIT
License (weights) MIT
Source https://github.com/microsoft/VibeVoice
Obligation MIT — attribution required if redistributing.
Responsible-use notice Microsoft's own docs state: "We do not recommend using VibeVoice in commercial or real-world applications without further testing and development... intended for research and development purposes only." This is not a license term — the MIT grant itself carries no field-of-use restriction — but it is a real business-risk signal (model maturity, misuse potential for synthetic voice) that AlienTech.Software should weigh deliberately before shipping this engine, not treat as a default green light.

3. UI Framework

Package Version License Notes
PySide6 >= 6.7.0 LGPL v3 (open-source edition) Qt6 Python bindings — UI, system tray, dialogs. LGPL requires end-users can relink against a modified Qt6 library. Satisfy by keeping Qt .dll files as separate files in the distribution (--onedir PyInstaller mode, not --onefile).

4. Python Runtime Libraries

Package Version License Purpose
pywin32 >= 306 (Windows only) PSF / MIT Win32 COM, clipboard, thread-input APIs
pynput >= 1.7.6 LGPL v3 Global hotkey listener
PyYAML >= 6.0 MIT stacks.yaml / user.yaml config parsing
numpy >= 1.24.0 BSD 3-Clause Audio buffer manipulation
sounddevice >= 0.4.6 MIT Cross-platform audio playback (ML engines)
soundfile >= 0.12.1 BSD 3-Clause WAV / FLAC read-write
lameenc >= 1.3.6 LGPL v2+ Pure-Python MP3 encoder; libmp3lame bundled in wheel
psutil >= 5.9.0 (perf extra) BSD 3-Clause System metrics in perf benchmarks; not in production build

LGPL packages (pynput, lameenc): Same dynamic-linking requirement as PySide6. Keep shared libraries replaceable in distributed builds.

5. ML Framework & Inference Stack

Lazy-imported — not loaded at startup; only activated when an ML stack is selected.

Package Version License Notes
PyTorch >= 2.0.0 BSD 3-Clause Core tensor / CUDA runtime
transformers >= 4.30.0 Apache 2.0 HuggingFace model loading and tokenizers
safetensors >= 0.4.0 Apache 2.0 Safe weight serialization format
accelerate >= 0.20.0 Apache 2.0 Device-aware model loading helpers
huggingface_hub >= 0.20.0 Apache 2.0 Model weight download and caching
onnxruntime Transitive (piper-tts) MIT ONNX inference runtime for Piper voices

6. TTS Engine Python Packages

Package Version License Source
kokoro >= 0.9.0 Apache 2.0 https://github.com/hexgrad/kokoro
piper-tts >= 1.2.0 MIT https://github.com/rhasspy/piper
dia Source (git+) Apache 2.0 https://github.com/nari-labs/dia
f5-tts Transitive / direct install MIT https://github.com/SWivid/F5-TTS
outetts Direct install MIT https://github.com/edwko/OuteTTS

7. Development & Test Tools (not distributed)

No distribution obligation applies to these packages.

Package License Purpose
pytest MIT Test runner
pytest-cov MIT Coverage measurement
ruff MIT Python linter and formatter
pyinstaller GPL + bootloader exception Freezes .exe; bootloader exception permits MIT-licensed output

8. External APIs & Cloud Services

AlienVox does not call any external APIs or cloud services at runtime.

  • All TTS inference is local and in-process — no network calls during speech.
  • Model weights are downloaded once from Hugging Face Hub (public, unauthenticated) and cached in %LOCALAPPDATA%\com.alientech.alienvox\.models\.
  • Telemetry is written to a local JSONL file only — no network transmission.

No API keys are required. No user data leaves the machine.

9. Maintaining This Document

When adding a new dependency:

  1. Add the package to python_app/requirements.txt with a version pin.
  2. Add a row to the appropriate section here, including: package name, version, SPDX license identifier, and source URL.
  3. If the license is LGPL or Apache 2.0, add or update the obligation note.
  4. For new ML models, fill in the full table in Section 2 — the weights license is separate from the code license and must be verified independently.