What problem does it solve?
StepFun's stepaudio-2.5-tts model replaced the old voice_label emotion tags with natural-language instruction parameters and inline () prosody directives, breaking existing step-tts-2 integrations and introducing stricter content censorship that blocks lines containing words like 死 or 消失. This Skill provides tested scripts and migration playbooks so you can synthesize emotional speech, batch-generate voice lines, and migrate legacy projects without hitting undocumented API errors.
Core Features & Use Cases
- Contextual Speech Synthesis: Generate Chinese/Japanese audio with global
instruction tone directives (≤200 chars) and inline () parentheses for mid-sentence pauses, whispers, and emotion pivots.
- Batch Voice Line Generation: Process JSONL files of game/app dialogue with per-line censorship handling that logs blocked IDs and continues instead of aborting.
- Migration Playbook: Complete step-tts-2 → stepaudio-2.5-tts guide with emotion-tag-to-instruction rewrite tables, non-destructive A/B directory strategy, and duration trade-off data (~20% longer output).
- Use Case: You maintain a visual novel with 30 Chinese voice lines on step-tts-2. Use this Skill to rewrite emotion tags as natural-language instructions, regenerate the corpus into a parallel
zh_v25/ directory, handle the 2 censored lines with fallback, and A/B compare durations before switching production.
Quick Start
Ask the AI to synthesize a Chinese voice line with a specific emotion using the stepfun-tts skill, for example: generate an mp3 of "你好，我是蕾格" with a warm, encouraging tone.
