| name | mofa-fm |
| description | TTS and voice cloning. Triggers: voice, TTS, text to speech, 语音, 播报, read aloud. |
MoFA FM — Text-to-Speech
One-line: synthesize speech (preset or cloned voices) via the local
ominix-api server. Background-spawning — call fm_tts directly with the
full text and the audio file is delivered automatically.
CRITICAL anti-patterns
Do NOT call fm_voice_list as a precheck before fm_tts or before
podcast generation. fm_voice_list is a recovery and explicit-listing
tool, not a TTS prerequisite. If a user names a voice you do not
recognise, call fm_tts directly with that voice — the error path is
short and recoverable, and a preemptive catalog dump wastes a turn and
can prevent the actual TTS call from happening.
Call fm_voice_list only when:
fm_tts or podcast generation returns Unknown voice 'NAME',
Voice not found, or an equivalent unavailable-voice error.- The user explicitly asks to list or browse available voices.
Do NOT use spawn, shell scripts, or manual text splitting around
fm_tts. The tool is spawn_only — call it once with the full text.
Do NOT use fm_tts for podcasts. Use podcast_generate from
mofa-podcast instead.
Quick reference
- Default voice:
vivian. Other presets and custom-voice save / list /
delete behaviour are documented in docs/full-guide.md.
- Style prompt (
prompt argument): leave empty for natural
content-aware tone; set it to override with a consistent style — see
the table in docs/full-guide.md.
- Voice cloning: save with
fm_voice_save (3-10s clear-speech clip),
then pass the same name as fm_tts's voice argument. Local-only
save — no server round-trip at save time.
