FAI Voice

{
  type: 'object',
  properties: {
    stt: {
      type: 'object',
      properties: {
        provider: { type: 'string', enum: ['azure-speech', 'whisper', 'google-stt', 'deepgram'], default: 'azure-speech' },
        language: { type: 'string', default: 'en-US' },
        continuous: { type: 'boolean', default: true },
        profanityFilter: { type: 'string', enum: ['none', 'masked', 'removed'], default: 'masked' },
        diarization: { type: 'boolean', default: false, description: 'Speaker identification for multi-party calls.' }
      },
      additionalProperties: false
    },
    llm: {
      type: 'object',
      properties: {
        provider: { type: 'string', default: 'azure-openai' },
        model: { type: 'string', default: 'gpt-4o' },
        streaming: { type: 'boolean', default: true },
        maxTokens: { type: 'integer', default: 500, description: 'Keep responses concise for voice.' },
        systemPrompt: { type: 'string', description: 'Agent persona and instructions.' }
      },
      additionalProperties: false
    },
    tts: {
      type: 'object',
      properties: {
        provider: { type: 'string', enum: ['azure-speech', 'elevenlabs', 'google-tts', 'openai-tts'], default: 'azure-speech' },
        voice: { type: 'string', default: 'en-US-JennyNeural' },
        speed: { type: 'number', minimum: 0.5, maximum: 2.0, default: 1.0 },
        style: { type: 'string', enum: ['neutral', 'cheerful', 'empathetic', 'professional'], default: 'professional' }
      },
      additionalProperties: false
    },
    latency: {
      type: 'object',
      properties: {
        sttTarget: { type: 'string', default: '< 1s', description: 'STT recognition latency target.' },
        llmTarget: { type: 'string', default: '< 2s', description: 'LLM first-token latency target.' },
        ttsTarget: { type: 'string', default: '< 1s', description: 'TTS synthesis latency target.' },
        endToEnd: { type: 'string', default: '< 3s', description: 'Total pipeline latency target.' }
      },
      additionalProperties: false
    },
    interruption: {
      type: 'object',
      properties: {
        handling: { type: 'string', enum: ['graceful', 'immediate-stop', 'queue', 'ignore'], default: 'graceful' },
        minSpeechMs: { type: 'integer', default: 500, description: 'Minimum speech duration before treating as interruption.' },
        silenceTimeoutMs: { type: 'integer', default: 2000, description: 'Silence duration before considering turn complete.' }
      },
      additionalProperties: false
    },
    recording: {
      type: 'object',
      properties: {
        enabled: { type: 'boolean', default: false },
        format: { type: 'string', enum: ['wav', 'mp3', 'ogg'], default: 'wav' },
        transcriptEnabled: { type: 'boolean', default: true },
        retention: { type: 'string', pattern: '^[0-9]+(d|m|y)$', default: '90d' }
      },
      additionalProperties: false
    }
  },
  additionalProperties: false
}

{
  "voice": {
    // See schema above for available options
  }
}

import { createSpecialties } from './engine/specialties/index.js';
const specs = createSpecialties(manifest);
// Access via: specs.voice