LOW-LATENCY LIVE LFM2.5-AUDIO

🎭 RoleForge: Live Voice NPC Director

Continuous WebAudio microphone, stable-prefix PCM streaming, bounded memory, and barge-in.

This build begins playback from guarded, proven-stable LFM prefix decodes and finishes with the clean reference tail. It automatically falls back to whole-response playback if prefix decoding fails. The microphone remains open and playback can be interrupted. Wear headphones to prevent echo-triggered interruption.

🎬 Server-side Director State

Private cue

Stored server-side cue: Hold steady
Not injected during official interleaved-contract validation.

Tool result

🎙️ Live Stage

IDLE

One click keeps the microphone open for the full session. No Record/Waiting toggle.

Choose a cue, wear headphones, then start the live session.

No completed model responses yet. Visitor speech is not separately transcribed.

Live diagnostics

Privacy: fictional speech only; never enter real secrets. Hidden game facts are enforced outside model context. No voice cloning, external inference API, WebRTC relay, or durable audio storage. Raw microphone/audio packets and conversation state are volatile and bounded to the live session.