Hung-yi Lee

New method improves fact-checking for computer-generated texts with ambiguous names.

AV-SUPERB evaluates audio and visual models across various tasks for better performance.

Reborn offers innovative solutions for automatic speech recognition without labeled data.

A new framework enhances language models by recognizing and responding to different speech styles.

A new system to evaluate audio codec performance across various applications.

A new framework for assessing foundation models in speech tasks.

A new tool to assess language models' continuous improvement through feedback.

New methods enhance speech recognition in noisy environments using adaptive techniques.

A new method improves machine dialogue through pseudo-stereo data.

DogeRM combines general and domain-specific models to enhance language model performance effectively.

A new framework enhances ASR performance using limited data and resources.

Examining how codecs retain emotional tones in voice data.

This article examines how user assistance can improve large language models' performance in generating SQL queries.

This article examines how structured generation affects language model reasoning and comprehension.

A new dataset highlights biases in speech models based on gender and age.

Research reveals how neurons in speech models recognize key features of sound.

This article discusses efficient training methods for speech models using self-supervised learning.

This study enhances emotion recognition systems for less common languages using high-resource data.

This study evaluates how well LLMs understand narrative tropes in movie summaries.

A look at the Codec-SUPERB challenge results and codec performance metrics.

New method for speech language models reduces need for extensive data.

Explore how Align-SLM is changing computer speech generation.

Creating an AI model for natural conversations in Taiwanese Mandarin.

A guide to effectively learning a new language with practical tips.

New methods improve how machines recognize spoken language.

New methods improve ASR systems for languages they haven't encountered before.

A new method merges models to improve safety and performance.