Feedback From Automatic Speech Recognition to Elicit Clear Speech in Healthy Speakers

Area of Science:

• Speech-Language Pathology
• Biomedical Engineering
• Acoustic Phonetics

Background:

• Clear speech is crucial for intelligibility, especially for individuals with speech disorders like dysarthria.
• Traditional methods for eliciting clear speech often require direct clinician feedback.
• Developing accessible, at-home practice methods is essential for improving speech outcomes.

Purpose of the Study:

• To evaluate the effectiveness of automatic speech recognition (ASR) feedback in eliciting clear speech in healthy young adults.
• To establish a preliminary understanding of ASR feedback's potential for individuals with dysarthria.
• To explore ASR as a tool for facilitating independent, at-home clear speech practice.

Main Methods:

• Twenty-three healthy adults produced speech in three conditions: habitual, clear, and ASR feedback.
• Mozilla DeepSpeech was used for real-time transcription accuracy feedback.
• Noise was introduced to ensure participants actively produced clearer speech for ASR.
• Acoustic measures and speech-language pathologist ratings assessed speech clarity and intelligibility.

Main Results:

• Speech produced with ASR feedback was clearer and more intelligible than habitual speech.
• Acoustic characteristics of ASR-elicited clear speech, such as vowel distinctiveness and speaking rate, resembled traditional clear speech.
• Individual speech features correlated differently with ASR accuracy, clarity, and intelligibility.

Conclusions:

• ASR feedback successfully elicited clear speech in healthy speakers, comparable to explicit clear speech instructions.
• The findings support ASR's potential as a tool for at-home speech therapy practice.
• Further research is warranted to investigate ASR feedback's efficacy in individuals with dysarthria.