Peer-reviewed research
Personalizing ASR for Dysarthric and Accented Speech with Limited Data
Joel Shor et al., Interspeech 2019, 2019
Read the originalRead on October 16, 2026
- What it is
- A 2019 Google study that fine-tuned a large speech recognizer on recordings from people with ALS and on accented speech from non-native English speakers. The authors measured how much word error rate fell and how much recorded speech the gain needed.
- How much weight it carries
- Peer-reviewed at Interspeech. The test set is a restricted vocabulary of read phrases, and the ALS group is 17 speakers rated FRS 3 or below, so the figures show what personalization can do rather than what a typical user will get. The DOI is the ISCA form and should be checked against the proceedings page.
Facts taken from this source
- The same 2019 study reported a 35 percent relative reduction in word error rate for accented speech after personalization.
“We train personalized models that achieve 62% and 35% relative WER improvement on these two groups” (Abstract)
- The personalized models in the 2019 Google study gave a 62 percent relative reduction in word error rate on speech from people with ALS.
“We train personalized models that achieve 62% and 35% relative WER improvement on these two groups” (Abstract)
- In the same 2019 study, 71 percent of the word error rate improvement on ALS speech came from only five minutes of training speech.
“We show that 71% of the improvement comes from only 5 minutes of training data.” (Abstract)
- In a 2019 Google study, personalized models brought word error rate on ALS speech down to 10 percent for mild dysarthria and 20 percent for more serious dysarthria.
“down to 10% for mild dysarthria and 20% for more serious dysarthria” (Abstract)
Pages that rest on this source
- How to Dictate Emails Faster Than You Type ThemGuide
- Speech Recognition and Accents: Accuracy and FixesGuide
- Speech Recognition for Stuttering and DysarthriaGuide
- Can dictation help non-native English speakers write work emails?Question
- Can you train dictation software to your accent?Question
- Word Error Rate (WER): How to Calculate and Read ItGuide
- Can voice recognition understand dysarthric speech?Question