An Automatic Prolongation Detection Approach in Continuous Speech With Robustness Against Speaking Rate Variations

Area of Science:

• Speech Pathology
• Computational Linguistics
• Signal Processing

Background:

• Automatic detection of speech prolongations is crucial for diagnosing and treating stuttering.
• Current methods primarily focus on diagnosis, with less attention to treatment support for slower speech.
• Speech-language pathologists (SLPs) require tools to assist in monitoring client progress during therapy.

Purpose of the Study:

• To develop an automated method for detecting speech prolongations to support SLPs in stuttering diagnosis and treatment.
• To provide a tool that helps monitor clients learning to speak more slowly during therapy sessions.

Main Methods:

• Speech signals were processed using Perceptual Linear Predictive (PLP) features.
• Correlation similarity measures were used to analyze successive speech frames.
• Prolonged segments were identified when highly similar frames exceeded a speaking rate-dependent threshold.

Main Results:

• The method achieved high detection accuracies of 99% on the UCLASS database and 97.1% on a Persian speech database.
• The system demonstrated robustness against variations in speaking rate (70-130% of normal).
• Performance was comparable to or better than three high-performance studies in automatic prolongation detection.

Conclusions:

• The proposed method offers an effective and robust approach for automatic prolongation detection in stuttering.
• This tool can significantly aid speech-language pathologists in both the diagnostic and therapeutic phases of stuttering intervention.
• Further research can explore integration into real-time clinical feedback systems.