The kitchen-sink TTS pipeline for your ebook pile

What it does

epub2tts ingests a DRM-free epub or text file and emits a standard M4B audiobook, complete with chapter markers, cover art, and sentence pauses. It wraps several TTS backends—Coqui (including XTTS voice cloning), Microsoft Edge, OpenAI, and Kokoro—and can resume generation if interrupted.

The interesting bit

The maintainer is admirably honest: the README opens by steering users toward newer, engine-specific forks for better Coqui cloning and long-form Microsoft voices, suggesting this repo has become the general-purpose proving ground while the spin-offs handle the heavy lifting.

Key highlights

• Outputs proper M4B files with automatic chapter detection and optional cover art.
• Supports five TTS engines: Coqui (VITS/XTTS), OpenAI, MS Edge, and Kokoro.
• Clones a speaker from 1–3 short audio clips via Coqui XTTS v2.
• Resumes where it left off, useful when local inference crawls at 0.15×–3.7× realtime.
• Parallel chapter rendering via --threads, though local models need tuning.

Caveats

• DRM-free epubs only.
• Kokoro currently ignores speed configuration; a fix is pending.
• Deepspeed may auto-detect but fail silently without a proper CUDA toolkit setup.

Verdict

Good for developers who want one tool to sample multiple TTS backends without writing their own pipeline. Skip it if you already know you want only Edge or only Coqui voice cloning—the author’s own spin-offs are likely sharper.