If you intended to convey a real-time S2S model, we currently use our own transcription service instead of the model.