speech-server.yaml.example

# speech-server configuration
# All fields are optional; omitted fields use the built-in defaults shown here.
# Config discovery order:
#   1. SPEECH_SERVER_CONFIG env var (path to this file)
#   2. ./speech-server.yaml in the working directory (this file, if present)
#   3. Built-in defaults
#
# speech-server.yaml is gitignored so local IPs stay out of version control.
# Copy this file to speech-server.yaml and edit as needed.

log_level: notice     # trace | debug | info | notice | warning | error | critical

servers:
  http:
    host: 127.0.0.1       # Bind address; override with HTTP_HOST env var or Vapor's --hostname flag.
    port: 8080            # Listening port; override with HTTP_PORT env var or Vapor's --port flag.
    upload_limit_mb: 500  # Maximum multipart upload size for /audio/transcriptions
  wyoming:
    host: 127.0.0.1       # Bind address for Wyoming TCP server; override with WYOMING_HOST env var.
    port: 10300           # TCP port for Wyoming protocol (Home Assistant). 0 = disabled; override with WYOMING_PORT env var.

stt:
  engine: parakeet      # Speech-to-text engine. Currently only: parakeet
  parakeet:
    model_version: v3   # v3 = Parakeet TDT 0.6B v3, multilingual (25 langs, default)
                        # v2 = Parakeet TDT 0.6B v2, English-only (higher recall)

tts:
  engine: pocket_tts    # Text-to-speech engine: pocket_tts | avspeech | kokoro

  # PocketTTS settings (only used when engine: pocket_tts)
  pocket_tts:
    sanitize_emoji: true  # Strip emoji before synthesis (default true)

  # AVSpeech settings (only used when engine: avspeech)
  # Uses macOS's built-in AVSpeechSynthesizer — no model downloads, 150+ voices.
  # avspeech:
  #   default_voice: Samantha   # Short name or full identifier; nil = system locale default
  #   sample_rate: 22050        # Native AVSpeech output rate (Hz); change only if needed

  # Kokoro TTS settings (only used when engine: kokoro)
  # Uses FluidAudio's Kokoro model — 50 voices across 8 languages, 24 kHz, high quality.
  # kokoro:
  #   default_voice: af_heart   # Any Kokoro voice ID (e.g. af_heart, am_adam); default af_heart