Knowing What to Stress: A Discourse-Conditioned Text-to-Speech Benchmark

2026-04-12

AI Summary

This work addresses the challenge that current text-to-speech (TTS) systems struggle to generate word-level stress aligned with pragmatic intent as dictated by discourse context. To bridge this gap, we introduce CAST—the first benchmark explicitly designed to systematically link discourse context with word-level stress. CAST employs carefully constructed contrastive context pairs and integrates language model–based stress prediction, TTS synthesis, and acoustic analysis to evaluate contextual awareness in stress realization. Experimental results reveal that state-of-the-art TTS systems significantly underperform language models in generating contextually appropriate prosodic stress. The project provides an open-source release of the full dataset and associated toolchain, establishing a new benchmark for pragmatic prosody modeling in TTS research.

Technology Category

• Natural Language Processing: Speech
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Abstract

Spoken meaning often depends not only on what is said, but also on which word is emphasized. The same sentence can convey correction, contrast, or clarification depending on where emphasis falls. Although modern text-to-speech (TTS) systems generate expressive speech, it remains unclear whether they infer contextually appropriate stress from discourse alone. To address this gap, we present Context-Aware Stress TTS (CAST), a benchmark for evaluating context-conditioned word-level stress in TTS. Items are defined as contrastive context pairs: identical sentences paired with distinct contexts requiring different stressed words. We evaluate state-of-the-art systems and find a consistent gap: text-only language models reliably recover the intended stress from context, yet TTS systems frequently fail to realize it in speech. We release the benchmark, evaluation framework, construction pipeline and a synthetic corpus to support future work on context-aware speech synthesis.

Problem

• text-to-speech
• word-level stress
• discourse context
• prosody
• context-aware synthesis

Innovation

• context-aware TTS
• word-level stress
• discourse conditioning
• prosody modeling
• speech synthesis benchmark