Carnegie Mellon Students Publish Breakthrough Speech Research at Premier IEEE Conference
A team of computer science sophomores from Carnegie Mellon University in Qatar has achieved a remarkable milestone in speech recognition technology. Their research paper, titled “LLM-based post-ASR error correction for disordered speech,” has been accepted by the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) – widely recognized as the world’s premier venue for signal processing research .
The student researchers – Hangyi Wen, Mikiyas Assefa, and Anas Semsayan – developed an innovative approach that utilizes Large Language Models (LLMs) to improve speech recognition accuracy, specifically focusing on disordered speech patterns . This work is particularly significant because it addresses the challenges faced by individuals with speech impairments who often struggle with conventional automatic speech recognition (ASR) systems.
The project originated in the Summer Undergraduate Research Apprenticeship (SURA) program during the students’ first year, demonstrating how early exposure to research can yield exceptional results . Under the guidance of faculty advisor Eduardo Feo Flushing, assistant teaching professor of computer science, these sophomores accomplished what is typically considered a significant achievement even for experienced researchers.
The team will present their findings at ICASSP 2026, scheduled to take place from May 4-8, 2026, in Barcelona, Spain . This conference brings together thousands of researchers, engineers, and industry leaders from around the world, making it an exceptional platform for these young researchers to showcase their work .
Key Characteristics of the Research
| Characteristic | Description |
|---|---|
| Innovative Approach | Uses Large Language Models (LLMs) for post-ASR error correction, specifically targeting disordered speech patterns rather than general speech recognition |
| Accessibility Focus | Addresses the critical need for better speech recognition technology for individuals with speech disorders who are often poorly served by conventional systems |
| Technical Methodology | The system analyzes speech patterns after initial ASR processing, applying LLM capabilities to correct errors and improve overall recognition accuracy |
| Premier Publication Venue | Acceptance at ICASSP, the flagship conference of the IEEE Signal Processing Society, represents exceptional quality and significance in the field |
| Undergraduate Achievement | Sophomore-level students accomplishing research typically associated with graduate students or professional researchers |
Causes and Contributing Factors
Institutional Factors
| Contributing Factor | Explanation |
|---|---|
| SURA Program | The Summer Undergraduate Research Apprenticeship program provided first-year students with structured research opportunities, allowing them to begin their investigative work early in their academic careers |
| Carnegie Mellon’s Research Culture | CMU’s emphasis on combining technical excellence with real-world applications created an environment where undergraduate research is encouraged and supported |
| Faculty Mentorship | Guidance from Eduardo Feo Flushing, assistant teaching professor of computer science, provided the students with expert direction while allowing them to take ownership of their project |
| Qatar Campus Resources | Carnegie Mellon University in Qatar offers the same rigorous education as the Pittsburgh campus while providing unique opportunities for international collaboration and perspective |
Technical Factors
| Contributing Factor | Explanation |
|---|---|
| Advances in LLM Technology | Recent breakthroughs in Large Language Models provided the foundational technology that made this post-ASR correction approach feasible |
| Growing Need for Accessibility | Increasing awareness of accessibility challenges in technology created demand for solutions addressing disordered speech recognition |
| Interdisciplinary Approach | Combining speech processing, machine learning, and accessibility research allowed for novel solutions to persistent problems |
