What problem does it solve?
Digital-human talking-head videos need expressive, correctly pronounced voiceover audio, but raw scripts sent directly to CosyVoice TTS produce flat or mispronounced speech. This Skill turns approved voiceover lines into fine-grained CosyVoice SSML segments with controlled rate, pitch, volume, pauses, and pronunciation fixes before calling voice.speech.
Core Features & Use Cases
- SSML Performance Design: Splits approved scripts into performable segments, each with a complete
<speak rate pitch volume> block and a per-segment style prompt.
- Pronunciation Correction: Uses
<say-as>, <phoneme>, and <sub> to fix digits, dates, amounts, polyphonic characters, abbreviations, and keyboard shortcuts.
- Strict Constraint Enforcement: Blocks invalid markup such as
<prosody>, MiniMax <#0.6#> pauses, emotion fields, and negative pitch values, then runs a self-check before a single voice.speech call.
- Use Case: Inside a video-director digital-human workflow, an approved product-explainer script is split into hook, selling-point, and CTA segments, each synthesized with distinct pacing and tone, then merged into one final audio track.
Quick Start
Within a video-director digital-human voiceover flow, invoke this skill to convert the approved script into CosyVoice SSML segments and synthesize the audio with one voice.speech call.
