We study how speech conveys meaning beyond words, and how insights from speech science can support more transparent, human-centered speech technologies.
SPEAK Lab brings linguistic analysis, cognitive explanation, empirical methods, and AI-informed engineering together to better understand human speech and improve speech agents.
To advance theory and practice at the intersection of speech phenomena, cognition, and speech technology.
Human speech carries rich information through prosody, timing, fluency, voice quality, and interactional feedback. We investigate these phenomena from linguistic and cognitive perspectives, then apply the findings to speech technologies that can process and produce speech more appropriately.
We examine acoustic, temporal, prosodic, and fluency-oriented features that shape interpretation.
We ask how speech behavior reflects planning, perception, learning, and interaction.
We use empirical evidence to inform perceptive and productive speech-technology pipelines.
We develop and evaluate technologies that respond to human speech in transparent, useful ways.
Ongoing projects examine listeners, synthetic speech, second-language speech, and real-time feedback in human-computer communication.
Testing how disfluencies shape listeners' interpretation of ongoing speech in Japanese.
Exploring how AI agents handle hesitation and how closely synthetic behavior resembles human behavior.
Investigating where pitch-accent difficulties arise for nonnative speakers and how ASR systems might address them.
Developing feedback systems in which on-screen avatars respond to features of a speaker's fluency.
SPEAK Lab is based at Waseda University's Nishi-Waseda Campus and works across speech science, perceptual computing, neurolinguistics, language acquisition, and speech technology.
Research interests include speech fluency, disfluency, prosody, corpus-based speech analysis, and the role of AI in language and speech technologies.
The first four items are shown by default. Use the button to show the complete publication list from the overview document.
From corpus studies of hesitation to speech-language models and AI-generated speech.
The lab's work combines linguistic description, perception experiments, speech technology evaluation, and ethical reflection on AI-generated speech.
Testing whether Japanese text-to-speech systems are sensitive to semantic information when realizing pitch accent.
Comparing linguistic features, automatic annotation, and human judgment in the analysis of epistemic stance.
Showing how transcript representation matters for LLM-driven analysis of spoken-language corpora.
Examining clause-boundary effects in disfluency perception using evidence from a speech language model.
Exploring jitter, shimmer, and harmonicity as supplementary acoustic correlates of second-language fluency.
Positioning non-pathological disfluency as an important issue for inclusive and fair speech AI.
Considering how ChatGPT and related tools can be framed as collaborators in language education.
Discussing ethical considerations raised by adding disfluency-like behavior to AI-generated speech.
Comparing pause and hesitation behavior using parallel corpora of English and Japanese presentation speeches.
Analyzing variation in jitter, shimmer, and intensity of filled pauses and their contexts.
Presenting real-time feedback on acoustic measures of second-language speech fluency.
Comparing filled pauses across scripted and non-scripted spontaneous speech contexts.
Introducing the Fluidity system for speech practice and second-language fluency development.
Investigating the structural signaling effect of silent and filled pauses.
Using corpus evidence to compare when speakers use filled pauses to fill pauses.
Comparing form and temporal characteristics of filled pauses across first- and second-language speech.
Connecting silent and filled pauses to speech planning in first- and second-language production.
Comparing second-language speech fluency ratings by native and nonnative listeners.
Studying temporal variables in first- and second-language speech and perceptions of fluency.
Examining contextual factors in the use of um and uh as differential delay markers.
Introducing a corpus for investigating first- and second-language speech performance.
Contact SPEAK Lab about collaboration, graduate research, speech corpus work, and human-centered speech technology projects.
SPEAK Lab welcomes conversations with students and researchers interested in speech science, cognition, speech technology, and explainable human-centered AI systems.