VoiceBridge Core

Unified “listen → understand → help you think → speak” engine for teaching, VRChat accessibility, and meetings. This repo currently contains Phase 0 scaffolding (types, config, logging, stubs) to unblock early integration.

Status

• Phase 0: core dataclasses, config loader, logging, profile samples, CLI smoke script.
• Not yet implemented: actual ASR/LLM/TTS calls, audio I/O, agent behaviors.

Quick start (Phase 0)

• Prereq: Python 3.10+.
• Install deps: python3 -m pip install -r requirements.txt
• Configure: cp config/settings.example.yml config/settings.yml and fill API keys/devices (optional at this phase).
• Smoke test: python3 cli/smoke.py --profile Teaching (warns if keys missing; verifies profiles load).

Config

• Settings: config/settings.yml (API keys, default audio devices, extras). Missing file is tolerated during Phase 0.
• Profiles: config/profiles/*.yml. Two examples exist: Teaching and VRChat. Key fields:
  • input_mode: "manual", "asr", or "manual+asr"
  • tts_backend, default_voice, output_device
  • reply_strategy: auto_suggest, auto_speak, max_suggestion_length, allow_agent_mode
  • prompts: mode-specific prompt templates (e.g., suggestion, explain, translate)

Repo layout (current)

• core/: types, config loader, logging helpers
• services/: abstract ASR/LLM/TTS interfaces
• conversation/: session manager
• understanding/: intent/suggestion stubs
• orchestrator/: flow coordinator shell
• audio_io/: audio input/output interfaces
• config/: settings and profile YAMLs
• cli/: smoke test entrypoint
• docs/architecture.md: full architecture and phase roadmap

Next steps

• Add concrete service implementations (OpenAI/Whisper/ElevenLabs) and wire to orchestrator.
• Implement text-only flow (intent + suggestions) in CLI.
• Extend audio I/O backends and agent auto-speak policy window.