Rule based methods produce more natural speech than prompting in voice simulators

Prompts versus Rules: Auditing and Controlling Speech Naturalness Behaviors in Voice User Simulators

Computation and Language

Summary

It can be hard to make computer programs that talk like real people, especially when they need to sound natural with pauses, interruptions, or back-and-forth noises. The authors looked at two ways to make these talking programs: using detailed instructions for language models, and using specific rules crafted by experts. They found that just asking language models to sound natural often doesn’t work well, but the rule-based method created speech that matched real human conversation much better. This means specialized rules or models are important for making realistic voice simulators.

What this means in practice

Authors

Riqiang Wang, Elena Khasanova, Harsh Saini, Lex Konnelly, Parsa Kavehzadeh, Matthias Lee, Mohamed Attia

Abstract

As voice agents gain more popularity commercially, the user simulators used to evaluate the deployed agents are also being developed to include more realistic, variable, and diverse speech naturalness behaviors -- disfluency, interruption and backchanneling. The quality of the user simulator directly affects the validity of agent evaluation results. However, we find that most studies so far have not examined in detail whether the intended configuration for these behaviors is realized in the simulation. In this study, we audit the realized naturalness behaviors of tau-Voice, our own LLM-based prompting approach across three models, and our rule-based injection algorithm for disfluency, interruption, and backchanneling. We find that prompting for these behaviors is unreliable and produces speech inconsistent with the instructions, placed and distributed less naturally than the instruction implies. In contrast, our rule-based, model-free algorithm produces controllable and diverse naturalness behaviors more aligned with natural speech. Our results suggest that LLMs not purpose-trained for user simulation are not sufficient on their own to represent authentic user behavior, and that linguistically informed deterministic approaches or specialized models are needed to close the gap; auditing and reporting realized naturalness behaviors, rather than configured settings, is what makes that gap visible.