# StutterNet: Stuttering Detection Using Time Delay Neural Network

**Authors:** Shakeel A. Sheikh, Md Sahidullah, Fabrice Hirsch, Slim Ouni  
**arXiv:** [2105.05599](https://arxiv.org/abs/2105.05599)  
**Date:** 2021-05-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2105.05599v2/

## Key points

- StutterNet is a novel approach to detecting stuttering using deep learning techniques
- It relies solely on the acoustic signal, unlike most existing methods that use automatic speech recognition (ASR) combined with language models for stuttering detection
- The system uses a time-delay neural network (TDNN) that captures contextual aspects of disfluent utterances
- StutterNet outperforms the state-of-the-art residual neural network based method when evaluated on the UCLASS stuttering dataset consisting of over 100 speakers
- The number of trainable parameters in StutterNet is substantially less due to the parameter sharing scheme of TDNN, making it an efficient and effective tool for detecting stuttering in real-world scenarios
- StutterNet represents an important advancement in the field of stuttering detection through its innovative use of deep learning techniques and reliance solely on acoustic signals
- It has significant potential for improving the accuracy and efficiency of stuttering detection in a wide range of applications.

## Summary

StutterNet is a novel approach to detecting stuttering using deep learning techniques. Unlike most existing methods in this domain, which rely on automatic speech recognition (ASR) combined with language models for stuttering detection, StutterNet relies solely on the acoustic signal. The system uses a time-delay neural network (TDNN) that is capable of capturing contextual aspects of disfluent utterances. The proposed method achieves promising results and outperforms the state-of-the-art residual neural network based method when evaluated on the UCLASS stuttering dataset consisting of over 100 speakers. Additionally, the number of trainable parameters in StutterNet is substantially less due to the parameter sharing scheme of TDNN. This makes it an efficient and effective tool for detecting stuttering in real-world scenarios. Overall, StutterNet represents an important advancement in the field of stuttering detection through its innovative use of deep learning techniques and reliance solely on acoustic signals. It has significant potential for improving the accuracy and efficiency of stuttering detection in a wide range of applications.