Short Drama Voice Studio

What it solves

Vertical short-drama episodes usually need dialogue audio that can go straight into an edit: one fixed voice per character, line order matching the script, and labeled clips that can be redone sentence by sentence. Generic TTS pipelines often mix character voices, fragment lines too aggressively, or return files without character IDs, line IDs, duration, and billing facts, making later replacement hard.

How it works

The skill assumes one hard input: a single episode script with speaker annotations. It does not treat bare character names as a script, and it does not invent dialogue. The core workflow is:

• Build a character table and line ledger: character ID, line ID, spoken text, and target file name.
• Use beatra.voices.list to choose voices, keeping the same character's voice_id fixed across the episode.
• Read the live text_to_speech information card, then confirm the character table, block count, weighted character count, and estimate before the first paid TTS call.
• Submit each speech block once with beatra.speech.synthesize, then poll it to completion with beatra.tasks.get.
• Deliver numbered clips in script order, including real duration, MIME type, URL or artifact, resolved model, and billing.net_charged_credits.

Voice cloning is only used when the user explicitly authorizes it. A suitable sample is usually 10 to 300 seconds, no larger than 20 MiB, and should be clean single-speaker audio. Insufficient balance should not trigger a clone request.

Scope and cautions

This skill fits vertical short-drama episode voice-over. Radio dramas, game dialogue packs, storyboard-to-video workflows, and personal cloned voice series belong to adjacent routes. Keep input under 50,000 characters, split by sentence or section, and avoid mid-sentence chunks. If the bundled scripts/mcp_client.py update fails, it fails open: the current command can still run, but an update failure is not a reason to retry paid generation.