scope:

This Recommendation ¹ describes a methodology on how to create a conversation simulation environment that is able to produce conversations that replicate certain characteristics of similar conversations recorded in a conversation test as described in [ITU-T P.800], [ITU-T P.805] and [ITU-T P.804]. The parameters aimed to replicate with the simulation approach are the turn-taking behaviours, the lengths of talk spurts, the interactivity of the conversation, as well as the intents conveyed in each utterance. In contrast to artificial conversational speech described in [ITU-T P.59], the simulation replicates the interactivity patterns of concrete conversation scenarios (e.g., conversation scenarios of [ITU-T P.805] Appendices IV-IX), as well as the impact of degradations such as transmission delay and packet loss on the conversation. The simulated conversations may be used in combination with existing quality models to predict the conversational quality but in itself does not predict the conversational quality.

¹ This Recommendation includes an electronic attachment with a reference implementation of the conversation simulation in Python (Annex A).