Fact · Accessibility and health
1.5hours of target speech
The 2019 system was adapted from a model trained on adult speech using only 1.5 hours of recordings of the target children.
Checked against its source on October 16, 2026
In the source's own words
“Working with only 1.5 hours of target data in which children perform the Clinical Evaluation of Language Fundamentals Recalling Sentences task”
Improving ASR Systems for Children with Autism and Language Impairment Using Domain-Focused DNN Transfer Techniques, Robert Gale et al., Interspeech 2019, ISCA Archive, 2019, Abstract.
Open the original- What it measures
- The method is called transfer learning: a large model trained on adult speech is adjusted with a small amount of recordings from the target group. The paper also added recordings of typically developing children to improve the result.
- What it does not show
- A small adaptation set means the result depends on the specific children and task, so it should not be generalized to all autistic speakers.