Voice Pipeline

CoreLayer includes a voice pipeline that lets you interact with Jarvis through speech.

Voice Pipeline Overview

Wake Word Detection
  → Speech Recognition (ASR)
  → Intent Processing (LLM)
  → Text-to-Speech (TTS)
  → Audio Output

  ↕ Barge-in interruption at any point

Configuration

{
  "voice": {
    "enabled": true,
    "wakeWord": {
      "enabled": true,
      "keyword": "jarvis"
    },
    "asr": {
      "provider": "web-speech",
      "language": "en-US"
    },
    "tts": {
      "provider": "mimo",
      "streaming": true,
      "sentenceLevel": true
    }
  }
}

Wake Word

Uses Picovoice Porcupine for wake word detection:

• Runs locally in the browser/webview
• Low CPU usage when idle
• Configurable keyword (default: "jarvis")

Speech Recognition (ASR)

Two providers available:

Text-to-Speech (TTS)

MiMo TTS is a cloud-based text-to-speech service that supports sentence-level streaming — it starts speaking each sentence as soon as it's generated, rather than waiting for the full response. This reduces perceived latency significantly.

Streaming TTS speaks responses as they're generated, reducing perceived latency.

Barge-in Interruption

When enabled, you can interrupt Jarvis mid-response by speaking. The pipeline will:

1. Stop the current TTS playback
2. Process your new input
3. Respond to the interruption

Voice Overlay

When voice is active, a floating overlay appears showing:

• Current state (listening, processing, speaking)
• Waveform visualization
• Transcription text

Next Steps

• Enable Voice Workflow — step-by-step setup
• Core Concepts: Jarvis — how voice integrates with the assistant