Baiyin AI Voice Generation

Problem Solved

When voice generation requirements hit concrete APIs, engineers can get stuck on field types, model identifiers, audio URLs, and parameter boundaries: an empty prompt can trigger 400, modelCode should not be hard-coded, params.speed_level and params.pitch_level must be numbers, and voice_url must be a publicly accessible http or https audio URL. Baiyin AI Voice Generation skill converges natural-language requests into stable Baiyin open-platform task requests, avoiding guessed models and fallback fields.

How It Works

• Runtime constraints: uses https://ai.hikoon.com, asks for BAIYIN_API_KEY only when it is missing.
• Model selection: fetches the available voice models, shows modelName to users, and submits the returned modelCode.
• Parameter mapping: maps expressions like “slower,” “higher pitch,” or “healing” to speed_level, pitch_level, and emotion; uses defaults when unspecified to reduce follow-up questions.
• Tone handling: silently fetches the tone list and defaults to the first audioUrl, or accepts a user-provided public audio URL for tone cloning.
• Task loop: reuses the latest taskId for status queries and returns the audio link only when the status is succeeded.

Boundaries and Notes

It fits voice synthesis, status lookup, and result retrieval on the Baiyin open platform. Keep prompt non-empty, obtain models from live endpoints, provide a public voice_url rather than local or file:// paths, and send integer values 1-5 for numeric parameters. For queued, processing, or failed statuses, return the status or backend error truthfully instead of fabricating completion.