Designation for the operating model of voicebots in which speech is converted into text via a speech recognition solution, analyzed by an LLM, and the response is subsequently generated by converting text back into speech via a speech generator. In addition to the two conversion steps, time is also required to determine the moment the speaker has finished speaking, so that the voicebot can process the input and formulate a response, for example by consulting a knowledge base. The various conversions require processing time, which is due to latency contributes. In addition, auditory information from the conversation, such as speed, pitch, and volume, is lost during the conversion to text.
See also speech-to-speech, voicebot, voice recognition, Voice activity detection