name: mofa-podcast
description: Multi-speaker podcast and dialogue generation with TTS voice cloning. Triggers: podcast, 播客, multi-speaker audio, 多人对话, 多人语音, radio show, talk show, 对话生成, 锵锵三人行, 两人对话, 三人对话, voice dialogue, 用XX的声音, 用XX和XX的风格, 声音模仿, 角色对话, 配音.
version: 0.4.7
author: hagency
always: false
requires_bin: mofa-podcast
pipeline: architecture.dot
MoFA Podcast — Multi-Speaker Podcast Generator
Generate professional multi-speaker podcasts from a topic or text. The pipeline
collects speaker preferences, expands content into a scripted dialogue with
emotion and music cues, lets you review before generation, then produces a
final MP3.
Flow at a glance
- Tell the agent your podcast topic (or paste source text).
- The agent will ask you for:
- Speakers (1–5): name + voice for each.
- Genre: drama, news, talk-show, interview, storytelling, debate, or custom.
- Length: target duration in minutes.
- A full script is generated in markdown with
[Speaker - voice, emotion] format.
- Review the script — approve, request edits, or cancel.
- On approval, the engine generates all voices and assembles the final MP3.
Anti-pattern: voice listing is not a precheck
Do NOT call fm_voice_list from mofa-fm as a precheck before generating a
podcast. If the user names speaker voices, use those names directly in the
script and let podcast_generate / TTS return a recoverable unknown-voice
error if a voice is unavailable. Use voice-listing tools (podcast_voices,
fm_voice_list) only when the user explicitly asks to browse or choose voices.
Read on demand
The reference material lives next to this file. Load only what the current
turn needs:
Output
Final artifacts land under skill-output/mofa-podcast/:
- Script:
script.md.
- Per-segment WAVs:
segments/*.wav.
- Final audio:
podcast_<timestamp>.mp3 (or .wav fallback when MP3
conversion is unavailable).
