Seshat TTS 1.0 packages the current Windows desktop build for OCR and local-model-assisted text-to-speech playback.
Highlights:
- Hotkey-driven capture from a selected monitor or window region.
- Dedicated stop-playback hotkey so active speech can be cancelled without starting another capture.
- Tesseract OCR text extraction with optional OpenAI-compatible local LLM cleanup.
- Optional local vision LLM mode that sends the selected region image directly to a vision-capable endpoint for text extraction.
- Pocket TTS playback through the uvx-server backend with stream cancellation on repeated reads.
- Built-in Pocket TTS voice selection plus custom WAV, MP3, and cached safetensors voice references.
- MP3 custom voice references are converted once into cached WAV prompts before Pocket TTS processes them.
- Packaged Windows EXE includes Seshat resources, Tesseract files when available on the build machine, uvx when found, and ffmpeg support for MP3 voice conversion.
Validation:
- Test suite: 13 passed.
- Source compile check passed.
Notes:
- Windows is the primary supported platform.
- First custom voice or Pocket TTS use may still download/cache model data in the user's normal cache directories.
- Third-party components retain their own licenses; see THIRD_PARTY_NOTICES.md.