A proof-of-concept study for automatic speech recognition to transcribe AAC speakers' speech from high-technology AAC...

Area of Science:

• Speech and Language Pathology
• Computer Science
• Assistive Technology

Background:

• Automatic Speech Recognition (ASR) is used for non-typical speech and language sample transcription.
• The use of ASR for high-tech Augmentative and Alternative Communication (AAC) systems remains unexplored.
• Language sample analysis is crucial in communication sciences and disorders.

Purpose of the Study:

• To investigate the feasibility of using an AAC-ASR model for transcribing language samples from high-tech AAC systems.
• To compare the transcription accuracy of a developed AAC-ASR model against established ASR models (CMU Sphinx, Google Speech-to-text).

Main Methods:

• Developed a proof-of-concept AAC-ASR model to transcribe simulated AAC speaker language samples.
• Evaluated the Word Error Rate (WER) of the AAC-ASR model using testing files.
• Compared the AAC-ASR model's WER with CMU Sphinx and Google Speech-to-text.

Main Results:

• The developed AAC-ASR model achieved a Word Error Rate (WER) of 28.6%.
• This significantly outperformed CMU Sphinx (70.7% WER) and Google Speech-to-text (86.2% WER) on the tested AAC speech samples.
• Demonstrated the feasibility of automated transcription for high-technology AAC-simulated language samples.

Conclusions:

• The AAC-ASR model shows promise for automatically transcribing high-technology AAC-generated language samples.
• This technology can support language sample analysis in communication sciences.
• Future research should focus on diverse AAC speech datasets and individual user speech pattern refinement.