Reproducible Speech Research With the Artificial Intelligence-Ready PERCEPT Corpora

Area of Science:

• Speech-language pathology
• Computational linguistics
• Developmental psychology

Background:

• Publicly available speech corpora enable reproducible research and clinical education.
• These datasets support perceptual training and the use of speech analysis tools.

Purpose of the Study:

• Introduce the PERCEPT (Perceptual Error Rating for the Clinical Evaluation of Phonetic Targets) corpora: PERCEPT-R (Rhotics) and PERCEPT-GFTA (Goldman-Fristoe Test of Articulation).
• Provide over 36 hours of speech audio data from individuals aged 6-24 with and without speech sound disorders, focusing on /ɹ/.
• Highlight PhonBank as the repository and demonstrate the use of Phon software for data analysis.

Main Methods:

• Collected and curated over 36 hours of speech audio data.
• Organized data into PERCEPT-R and PERCEPT-GFTA corpora.
• Utilized PhonBank for data repository and Phon software for analysis.

Main Results:

• The PERCEPT corpora contain over 125,000 syllable, word, and phrase utterances.
• Demonstrated the utility of PERCEPT-R with Phon software through a research example.
• Established a Slack channel for user support and future data release information.

Conclusions:

• PERCEPT corpora, PhonBank, and Phon facilitate clinical training and research in child speech.
• These resources can enhance the reproducibility of studies on speech development and disorders.
• Potential exists for training artificial intelligence clinical speech technology for children with speech sound disorders.