Summary
TiCo is a simple post-training method that enables spoken dialogue models (SDMs) to follow time-constrained instructions and generate responses with controllable duration. Although existing SDMs can produce natural spoken responses, they lack time awareness and frequently fail at duration-related requests such as generating a response lasting about 15 seconds, as shown by an empirical evaluation of both open-source and commercial models. TiCo addresses this by having models estimate elapsed speaking time during generation using speech-time markers (STMs), such as <10.6 seconds>, which helps the model maintain temporal awareness and adjust remaining content to reach a target duration. The method is efficient and lightweight: it requires only a small amount of data, needs no extra question-answer pairs, and relies on self-generated data and reinforcement learning. This capability is valuable for real-world spoken systems like voice assistants and interactive agents, where controlling response length improves interaction quality. Paper by Kai-Wei Chang, Wei-Chih Chen, En-Pei Hu, Hung-yi Lee, and James Glass, published March 23, 2026 on arXiv (2603.22267).
Paper Overview
Field: NLP
Authors: Kai-Wei Chang, Wei-Chih Chen, En-Pei Hu, Hung-yi Lee, James Glass
Published: 2026-03-23
arXiv: 2603.22267
Abstract (translated)
We propose TiCo, a simple post-training method for enabling spoken dialogue models (SDMs) to follow time-constrained instructions and generate responses with controllable duration. This capability is valuable for real-world spoken language systems such as voice assistants and interactive agents, where controlling response duration can improve interaction quality. However, despite their strong ability to generate natural spoken responses, existing models lack time awareness and struggle to follow duration-related instructions (e.g., "Please generate a response lasting about 15 seconds"). Through an empirical evaluation of both open-source and commercial SDMs, we show that they frequently fail to satisfy such time-control requirements. TiCo addresses this limitation by enabling models to estimate the elapsed speaking time during generation through speech-time markers (STM), such as <10.6 seconds>. These markers help the model maintain time awareness and adjust the remaining content to reach the target duration. TiCo is simple and efficient: it requires only a small amount of data and no additional question-answer pairs, relying instead on self-generated data and reinforcement learning.
---
*Auto-collected on 2026-03-25.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177169037