Peer-reviewed research
Benchmarking Commercial Speech Recognition and Multimodal Large Language Models on Dysarthric Speech
Ali Alsayegh and Tariq Masood, International Journal of Intelligent Systems, 2026 · DOI 10.1155/int/6065038
Read the originalRead on October 16, 2026
- What it is
- A 2026 study from the University of Strathclyde that tested eight commercial speech-to-text services, including Whisper large-v3 and several Gemini and GPT-4o models, on the TORGO dysarthric speech corpus, reporting word error rate by severity.
- How much weight it carries
- Peer-reviewed and recent. It uses only 8 speakers from TORGO, with 2 mild, 2 moderate and 4 severe, recorded on a head-mounted microphone and tested once in September 2025 with default settings. Cloud services change often, so the numbers describe that window.
Facts taken from this source
- In a 2026 study of commercial systems, the leading services reached about 1 to 2 percent word error rate on mild dysarthria.
“mild dysarthria reached low single-digit WER, around 1% to 2% for the leading systems” (Abstract)
- In the same 2026 study, every commercial system exceeded 51 percent word error rate on severe dysarthria.
“severe dysarthria exceeded 51% WER for every system” (Abstract)
- A 2026 study notes that automatic speech recognition reaches word error rates below 5 percent on typical speech, but performance degrades sharply for dysarthric speakers.
“achieves word error rates (WER) below 5% on typical speech” (Abstract)
Pages that rest on this source
- Speech Recognition for Stuttering and DysarthriaGuide
- Can voice recognition understand dysarthric speech?Question
- What is a good word error rate?Question
- What is the most accurate speech-to-text model?Question
- Speech-to-Text APIs Compared: Price, Accuracy, SpeedGuide
- Word Error Rate (WER): How to Calculate and Read ItGuide
- Speech Recognition Benchmarks and LeaderboardsGuide
- How accurate is voice recognition software?Question
- Which speech recognition apps work for atypical speech?Question
- Which speech-to-text API is the best?Question