mimo-lecture-audio-skill

name	mimo-lecture-audio-skill
description	Generate MiMo-powered lecture narration audio from teaching notes, with optional ASR QA, subtitles, HTML review pages, WAV merge, duration stats, packaging, and runtime official-doc sync checking. Use progressive disclosure: route to the smallest script needed for the user's task.

MiMo Lecture Audio Skill

Use this Skill when the user wants to turn lecture notes, scripts, Markdown, HTML, or segmented narration text into MiMo audio. The core capability is MiMo TTS. ASR, subtitles, HTML player, audio merge, duration stats, and ZIP packaging are optional add-ons.

Progressive disclosure rule

Do not run the full pipeline by default. Choose the smallest script that satisfies the user's request.

• Only generate narration audio: use scripts/mimo_tts_batch.py.
• Transcribe generated audio for checking: use scripts/mimo_asr_transcribe.py.
• Generate subtitles for video/voiceover: use scripts/generate_srt.py.
• Generate a standalone review page: use scripts/generate_html_player.py.
• Add audio controls to an existing lecture HTML: use scripts/inject_audio_to_html.py.
• Merge segment WAV files: use scripts/merge_wav.py.
• Measure audio durations: use scripts/audio_duration.py.
• Check whether local rules match official MiMo docs and /v1/models: use scripts/check_official_docs.py.
• Run a common multi-step workflow: use scripts/run_pipeline.py with explicit flags.

After TTS synthesis completes, always execute the Post-TTS ask rule (see below), even if the user only asked for "generate audio".

Core workflow

1. Clean source notes if needed.
2. Create or receive segments.json.
3. Validate segments.
4. Run TTS.
5. Optionally run ASR, duration stats, subtitles, HTML player, merge, or ZIP.

Key MiMo API rules

• TTS uses /v1/chat/completions.
• Put narration text in assistant.content.
• Put style, tone, or voice design instructions in user.content.
• Use mimo-v2.5-tts for preset voices.
• Use mimo-v2.5-tts-voicedesign with voice_design_prompt; do not pass preset voice.
• Use mimo-v2.5-tts-voiceclone with voice_sample_path or data:audio/...;base64,....
• Use --stream only for preset TTS by default. Use --stream-all only when compatibility streaming is explicitly desired.
• ASR uses mimo-v2.5-asr and input_audio.data data URLs.

Self-update guard

This skill checks itself against official MiMo docs before synthesis. Two complementary layers:

• API layer: GET /v1/models — verifies the 4 models this skill uses are still listed. Catches deprecations (the V2 series was retired 2026.6.30).
• Doc layer: fetches llms.txt → .md sources and diffs preset voices / languages / formats against templates/known_rules.json.

Behavior: default non-blocking (read cache → warn → continue). Use --check-docs to make it blocking. Use --skip-check to bypass. See docs/self_update.md for details.

Default commands

Validate segments:

python scripts/validate_segments.py --segments output/segments.json

Generate TTS audio:

python scripts/mimo_tts_batch.py \
  --config templates/config.example.json \
  --segments output/segments.json \
  --out-dir output/audio \
  --manifest output/audio_manifest.json \
  --include-text-in-manifest

Run optional pipeline with explicit switches:

python scripts/run_pipeline.py \
  --input lecture.md \
  --tts \
  --duration \
  --html-player

Merge existing segment WAVs into one long audio:

python scripts/merge_wav.py \
  --manifest output/audio_manifest.json \
  --output output/full_course.wav

Post-TTS ask rule

After TTS synthesis completes (mimo_tts_batch.py or run_pipeline.py --tts), the agent must use AskUserQuestion to ask the user which post-processing steps they want. This is a multiSelect question with these options:

Option label	What it does	Script to call if selected
合并成一个长音频	Merge all segments into one WAV	merge_wav.py --manifest <manifest> --output <out_dir>/full_course.wav
生成 HTML 播放页	Standaway review page with audio players	generate_html_player.py --manifest <manifest> --output <out_dir>/player.html --title <title>
测各段时长	Measure each segment's duration	audio_duration.py --manifest <manifest> --update-manifest
生成 SRT 字幕	Subtitles for video/voiceover	generate_srt.py --manifest <manifest> --segments <segments> --srt <out_dir>/subtitles.srt
都不用，音频够用了	Skip all post-processing	Call nothing

Rules:

• If the user selects multiple items, execute in dependency order: duration → merge → srt → html-player (duration enriches the manifest so srt/html-player can show timings).
• After the user picks, call the scripts immediately via Bash. Do not ask again.
• If running via run_pipeline.py with flags like --merge --html-player already set, skip the ask (user already chose).

Optional pipeline flags

• --tts: generate MiMo TTS audio.
• --asr-check: transcribe generated audio for QA.
• --srt: generate SRT subtitles.
• --vtt: generate WebVTT subtitles.
• --html-player: generate standalone audio review page.
• --inject-html PATH: inject audio controls into an existing HTML lecture.
• --merge: merge WAV segments into full_course.wav.
• --duration: measure WAV duration and enrich manifest.
• --zip: package output directory.

Output expectations

For TTS tasks, return links or paths to generated audio files and audio_manifest.json. For optional add-ons, return only the requested artifacts, such as player.html, course.srt, course.vtt, full_course.wav, or the ZIP package.

Intermediate cleanup (optional)

The pipeline leaves several internal state files that most users do not need after a successful run: work/ (cleaned text + segments), duration_manifest.json, asr_manifest.json, merge_manifest.json, and *.jsonl logs.

Set the env var to auto-clean these after success:

export MIMO_AUDIO_CLEAN_INTERMEDIATES=1

When enabled, the pipeline deletes only the items above. It preserves all final deliverables: audio/*.wav, audio_manifest.json, full_course.wav, player.html, subtitles/*.srt, subtitles/*.vtt, asr/*.txt, <stem>_with_audio.html, and *.zip. Failed runs are never cleaned (so you can debug). To make this permanent on Windows, set it as a user environment variable.

Details

Load deeper docs only when needed:

• docs/tts.md
• docs/asr.md
• docs/subtitles.md
• docs/html_player.md
• docs/pipeline.md
• docs/mimo_v25_audio_rules.md
• docs/self_update.md
• docs/troubleshooting.md