Api:
  # Reuse one SpeechSynthesizer across requests instead of reloading ONNX each call.
  reuse_tts: true