Doc no. SK-404DatasheetRev 2026.11Sheet 2.11 of 4
Odysseus Speech Tools
Glue and tools for a local, OpenAI-compatible speech engine in Docker, built for Odysseus: Kokoro-82M speech, Faster-Whisper transcription and document-to-audio scripts.
Active
Source on GitHub
Built for Odysseus; Runs on Speaches
$ ./narrator.py test_docs/sample.md -o narration.mp3 --play
Block diagram
Characteristics
| Text-to-speech | Kokoro-82M (ONNX): neural voices with natural prosody |
|---|---|
| Speech-to-text | Faster-Whisper (Systran/faster-whisper-small by default), local and memory-efficient |
| Engine | The open-source Speaches server in a Docker container: an OpenAI-compatible API, CPU only |
| Narrator | narrator.py reads PDF, Markdown, EPUB and text aloud to MP3 |
| Audiobooks | pdf2epub_tts.py strips headers, footers, page numbers and citations, then builds a chaptered .m4b and a clean EPUB |
| Verification | verify_all.py and test_audio_engine.py check the engine end to end |
Overview & build notes
Connects Odysseus to a local Speaches Docker container serving OpenAI-compatible /v1/audio/speech and /v1/audio/transcriptions endpoints (http://localhost:8000/v1) powered by Kokoro-82M (ONNX) and Faster-Whisper.
Alongside real-time agent voice replies, the repository includes narrator.py for quick PDF/Markdown/EPUB-to-MP3 readout and pdf2epub_tts.py (which evolved into Kokoro Audiobook Studio) for chaptered .m4b conversion.
Both verify_all.py and test_audio_engine.py validate live model aliases, TTS/STT round-trips and container health end to end.
Files & artifacts
| Kind | File | Details | Size |
|---|---|---|---|
| Script | narrator.py | Document-to-MP3 CLI reader supporting PDF, Markdown, EPUB and plain text | GitHub |
| Script | pdf2epub_tts.py | Layout cleaner and chaptered .m4b / EPUB generator on GitHub | GitHub |
| Tests | verify_all.py | End-to-end TTS/STT engine verification suite on GitHub | GitHub |
| Docs | PDF2EPUB_TTS_README.md | Usage guide for document cleanup and multi-chapter audiobook builds | GitHub |