dsh-audiobook: Turn Books into Speech Inside DeepSeek Harness
aicivilization/dsh-audiobook
A DSH plugin that imports EPUB, PDF, TXT and more, then reads them aloud with Edge neural voices or offline Kokoro while remembering your listening position.
dsh-audiobook brings listening to DeepSeek Harness by turning EPUB, PDF, TXT, DOCX and Markdown files into narrated content you can follow along with. It solves the problem of e-books that can only be read, splitting chapters automatically, reading paragraph by paragraph and remembering your position. Narration can use device voices, Microsoft Edge neural voices or the offline Kokoro model, with speed from 0.75× to 2.5×, plus a sleep timer and headphone controls.
How to Install
dsh plugin --profile web add dsh-audiobook- Category
- Tools & Capabilities
- Platform
- DSH Plugin
- Author
- aicivilization
- Distribution
- Plugin
dsh-audiobook Key Features
dsh-audiobook Repo Summary
What Does It Do?
Sou Shu Ting is a DSH plugin for DeepSeek Harness that turns any book into an audiobook inside the harness. It solves the problem of having e-books you can only read, letting you import a file, have it split into chapters and spoken aloud in the sidebar, and resume exactly where you stopped. Narration can use Microsoft Edge neural voices, an offline Kokoro model, or the voices built into your device, with playback speed from 0.75× to 2.5×. It is maintained by AIcivilization under the MIT license.
Core Features
- Multi-format import with chapter detection: reads TXT with automatic encoding detection (UTF-8, UTF-16, GBK, Big5), EPUB in spine order with table-of-contents titles as chapter names, PDF text layers with line breaks rejoined into paragraphs, plus DOCX, Markdown and HTML; plain text is split on headings like "Chapter N" or falls back to roughly 8,000-character parts.
- Three narration engines: device voices run fully locally so text never leaves the machine; Microsoft Edge offers 12 Chinese and 6 English voices and sounds the most natural but requires a network connection; the offline Kokoro Chinese model provides 55 female and 45 male voices with a one-time download of about 147 MB.
- Paragraph-level tracking and jumping: text is cut into roughly 20–30 second segments, the current segment is highlighted, and clicking any segment starts playback there, with chapter navigation and a draggable whole-book progress bar.
- Persistent progress: position is saved after every segment both on the server and locally, so closing the panel or switching devices resumes from the same spot; re-importing the same book from another location is still recognized as the same title.
- AI tools inside the conversation: the plugin registers 4 tools so you can ask DeepSeek Harness in natural language to import a book, list what you are listening to, retrieve recently heard text or search the whole book, and jump to a chapter.
- Sleep timer and media controls: stop after 15 minutes up to 1.5 hours or at the end of the current chapter, with lock-screen and headphone controls for play, pause and segment skipping.
How to Use This Plugin?
After enabling it in DSH, open the right sidebar on the chat page and pick the reading tab, then drag a book into the panel, choose a file, or type a local path to import it, and select a voice and speed to start playback. You can also simply say things like "import ~/Downloads/book.epub" or "jump to chapter ten" in the conversation and let the registered AI tools handle it. The panel also opens in a mobile browser, and a DSH instance on a VPS works the same way; if ffmpeg is present, local speech is compressed to MP3 for smoother remote listening. Books, progress and settings live in a local data directory and are kept even if the plugin is removed.
How to Troubleshoot This Plugin?
- Nothing is read after import: DRM-protected EPUB files and scanned PDFs without a text layer cannot be narrated; use a DRM-free file or one with an extractable text layer.
- Edge voices stop working: this mode needs a network connection and an open-source speech component, and it is not an official Microsoft API, so it may break; switch to device voices or the offline model.
- Local speech takes long to start: Kokoro synthesis is slow, with the first segment taking over ten seconds before later segments are pre-synthesized during playback; lower the worker count per the official instructions when memory is tight.
- Playback stops when the phone locks: device-voice narration may pause on lock screen, so use a server-side synthesis mode to keep listening.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: aicivilization/dsh-audiobook. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.