Millions of people are affected by stuttering and other speech disfluencies, with the majority of the world having experienced mild stutters while communicating under stressful conditions. While there has been much research in the field of automatic speech recognition and language models, stutter detection and recognition has not received as much attention. To this end, we propose an end-to-end deep neural network, FluentNet, capable of detecting a number of different stutter types. FluentNet consists of a Squeeze-and-Excitation Residual convolutional neural network which facilitate the learning of strong spectral frame-level representations, followed by a set of bidirectional long short-term memory layers that aid in learning effective temporal relationships. Lastly, FluentNet uses an attention mechanism to focus on the important parts of speech to obtain a better performance. We perform a number of different experiments, comparisons, and ablation studies to evaluate our model. Our model achieves state-of-the-art results by outperforming other solutions in the field on the publicly available UCLASS dataset. Additionally, we present LibriStutter: a stuttered speech dataset based on the public LibriSpeech dataset with synthesized stutters. We also evaluate FluentNet on this dataset, showing the strong performance of our model versus a number of baseline and state-of-the-art techniques.

FluentNet: End-to-End Detection of Stuttered Speech Disfluencies With Deep Learning

The presented paper introduces principal component analysis application for dimensionality reduction of variables describing speech signal and applicability of obtained results for the disturbed and fluent speech recognition process. A set of fluent speech signals and three speech disturbances—blocks before words starting with plosives, syllable repetitions, and sound-initial prolongations—was transformed using principal component analysis. The result was a model containing four principal components describing analysed utterances. Distances between standardised original variables and elements of the observation matrix in a new system of coordinates were calculated and then applied in the recognition process. As a classifying algorithm, the multilayer perceptron network was used. Achieved results were compared with outcomes from previous experiments where speech samples were parameterised with the Kohonen network application. The classifying network achieved overall accuracy at 76% (from 50% to 91%, depending on the dysfluency type).

/pdf/artificial-neural-networks-combined-with-the-principal-1d6ewci0.pdf

Artificial Neural Networks Combined with the Principal Component Analysis for Non-Fluent Speech Recognition

Strong presentation skills are valuable and sought-after in workplace and classroom environments alike. Of the possible improvements to vocal presentations, disfluencies and stutters in particular remain one of the most common and prominent factors of someone's demonstration. Millions of people are affected by stuttering and other speech disfluencies, with the majority of the world having experienced mild stutters while communicating under stressful conditions. While there has been much research in the field of automatic speech recognition and language models, there lacks the sufficient body of work when it comes to disfluency detection and recognition. To this end, we propose an end-to-end deep neural network, FluentNet, capable of detecting a number of different disfluency types. FluentNet consists of a Squeeze-and-Excitation Residual convolutional neural network which facilitate the learning of strong spectral frame-level representations, followed by a set of bidirectional long short-term memory layers that aid in learning effective temporal relationships. Lastly, FluentNet uses an attention mechanism to focus on the important parts of speech to obtain a better performance. We perform a number of different experiments, comparisons, and ablation studies to evaluate our model. Our model achieves state-of-the-art results by outperforming other solutions in the field on the publicly available UCLASS dataset. Additionally, we present LibriStutter: a disfluency dataset based on the public LibriSpeech dataset with synthesized stutters. We also evaluate FluentNet on this dataset, showing the strong performance of our model versus a number of benchmark techniques.

FluentNet: End-to-End Detection of Speech Disfluency with Deep Learning.

Stuttering is a neurodevelopmental speech disorder wherein people suffer from disfluency in speech generation. Recent research has applied machine learning and deep learning approaches to stuttering disfluency recognition and classification. However, these studies have focussed on small datasets, generated by a limited number of speakers and within specific tasks, such as reading. This paper rigorously investigates the effective use of eight well-known machine learning classifiers, on two publicly available datasets (FluencyBank and SEP-28k) to automatically detect stuttering disfluency using multiple objective metrics, i.e. prediction accuracy, recall, precision, F1-score, and AUC measures. Our experimental results on the two datasets show that the Random Forest classifier achieves the best performance, with an accuracy of 50.3% and 50.35%, a recall of 50% and 42%, a precision of 42% and 46%, and an F1 score of 42% and 34%, against the FluencyBank and SEP-28K datasets, respectively. Moreover, we show that the machine learning-based approaches may not be effective in accurate stuttering disfluency evaluation, due to diverse variations in speech rate, and differences in vocal tracts between children and adults. We argue that the use of deep learning approaches and Automatic Speech Recognition (ASR) with language models may improve outcomes, specifically for large scale and imbalanced datasets.

Stuttering Disfluency Detection Using Machine Learning Approaches

the purpose of this paper is to develop a speaker recognize system using Mel Frequency Cepstral Coefficients (MFCC) and Support Vector Machine (SVM). Today’s speakers recognize system are developed using various other techniques which results quite efficient. The methods proposed in this paper are producing fine results which should not be went unseen. First we are using MFCC which is used to extract features from the input provided to it where input is in the form of wav file. Secondly, after extraction of features SVM comes in the phase which does the matching of features extracted by MFCC and gives the recognized speaker name. Also it shows the graph of matched users. In this proposed method there are two divisions of system named training and testing. Training part comes under developer where developer train the system with the audio of authorized speakers and on that basis a database is build up. Testing gives the result back by authenticating the authorized user to get the system. In this way this system can be used to develop an authentication and forensic speaker recognition system.

Speaker Recognition using MFCC and Support Vector Machine

Disability in speech concerns many other communication problems such as hearing, and fluency. Stuttering is a neurodevelopmental disorder identified by the existence of dysfluencies during speech production. The disruptions of speech flow in stuttering have led a large body of researchers to examine the potential processes underlying speech-language production in people who stutter. This paper presents a survey on techniques of feature extraction and classification which are applied for automatic speech recognition, and also presents a comparative analysis of stuttering techniques on the basis of accuracy, sensitivity, specificity, and dataset size.

Shweta Khara

Papers

A Comparative Study of the Techniques for Feature Extraction and Classification in Stuttering