MiMo TTS and ASR Voice Toolkit

Problem

Voice synthesis in local scripts, content pipelines, or audiobook-style generation often hits three gaps: a raw TTS API can be unstable on long text, there is no fallback when a key is missing or rate limits are triggered, and teams still have to hand-roll batch files, numeric normalization, and output validation. MiMo TTS & ASR targets these workflow gaps by wrapping model calls into reusable command-line behavior.

How it works

The skill centers on scripts/tts.py and scripts/asr.py:
- Three TTS modes: preset voices, natural-language voice design, and reference-audio cloning through one entry point;
- Engineering fallbacks: falls back to edge-tts when MIMO_API_KEY is absent, and retries 429, 502/503, and timeout cases with randomized delay;
- Long-text handling: segmenting by punctuation, reading batch files, and watching directories for new .txt inputs;
- Output safety: inferring wav/mp3/ogg, checking empty or overly short files, and supporting cache, denoise, and normalization.

Boundaries

Cloud ASR remains reserved; local recognition usually needs a GPU with CUDA. The three TTS models are currently limited-time free, so keeping --cache is useful once billing starts. The skill depends on ffmpeg/ffprobe and python3, and fits content batch processing, audio demos, and speech prototyping rather than replacing official API authorization, quota, or compliance controls.