listen_book | LLM 网络安全

Turns dialogue-rich chapter Markdown into publishable audiobook MP4: multi-voice TTS, timed subtitles, character collage visuals, optional BGM, driven from a local web console.

Problem

Audiobook-style output needs consistent multi-voice audio, subtitles, and visuals—expensive to edit by hand.
Many chapters and frequent iterations require a repeatable batch pipeline, not one-off scripts.

Approach

Input: chapter MD (narration + dialogue); per-novel -id JSON config for paths and roles.
Pipeline: Edge TTS per role → SRT → looping collage background → FFmpeg chapter MP4 → optional BGM.
Surface: local FastAPI web UI; CLI for placeholders, LLM tagging, collage-only runs, etc.

Stack

TTS: edge-tts
Video: FFmpeg
Optional tagging: OpenAI-compatible API
UI: FastAPI / uvicorn
Python 3.10+

Outcomes

End-to-end MD → MP4 per chapter, including the web path.
Configurable voices via role gender; text normalization (e.g. illustration markers).
Pipeline architecture diagram (static, no app source)

UI & sample output

Three console tabs plus a sample frame (click to enlarge). Portfolio display only.

Web console: chapter Markdown editor with dialogue role chips — Tab 1 — Chapter MD, dialogue tags, regenerate MP4 from the editor.

Collage banner generation for video background — Tab 2 — Character standees, collage banner for loop background.

Character list with collage vs dialogue-only roles — Role grid: collage vs dialogue-only, upload and ordering.

Video generation tab with batch job log — Tab 3 — Single-chapter render, BGM, job log (FFmpeg + scripts).

Sample output frame with subtitles and character collage — Output — Multi-voice audio, timed subtitles, collage background.

Source & contact