Dialogue audio

Produce a multi-speaker conversation as one audio file: a podcast snippet, character dialogue, an interview, or an explainer back-and-forth. The lever is inline <|speaker:N|> tags in a single text input mapped to a speech.voices array, so both sides render in one call with automatic turn-taking. For one voice reading straight through, that is voiceover, not this.

Inputs to collect

• The script, written as alternating turns for two speakers. (Ask for it if the user only gave a topic.)
• The two voices (voice model IDs) and which speaker is index 0 vs 1. If none specified, pick two that contrast in pitch/cadence.
• Per-turn emotion, if any (excited, surprised, whispering). Optional, tag only the turns that need it.
• Optional: language (the model auto-detects and supports 80+) and output format.

Models

• Default: Fish Audio S2.1 Pro (fishaudio:s2.1@pro) - the one model here that renders two speakers in a single audioInference call via inline speaker tags, with per-speaker emotion control. Capability io:text-to-audio, status live.
• This is a model-specific feature, not a generic TTS one. Other text-to-audio models do one voice per request and would need separate calls plus downstream stitching.
• Confirm the live model + its schema via the runware-models + runware-run skills before calling - never hardcode a stale choice.

Workflow