Automatic identification and classification of stuttering in speech could significantly improve objective and consistent assessment of stuttered speech. Conventionally, assessment of stuttering in ...speech is done by counting number of dysfluent words as a proportion to the total number of words in a passage. It is also done by measuring the time of dysfluencies comparing with the duration of entire passage. However, its time consuming and results in destitute agreement with different individuals assessing the same speech data. Therefore, automatic recognition of stuttering in speech is used to automate the dysfluency count and type of dysfluency classification, thus providing consistent and objective assessment of stuttering in speech. This method can support Speech-Language Pathologists (SLPs) by doing tedious monotonous work and allowing more time for therapeutic sessions between SLPs and stutterers. This research paper explores the use of Artificial Fish Swarm method (AFSA) to optimize recognition and classification of stuttered events. AFSA is an optimization method inspired by the nature of collective fish movement and social behavior in search of food. In the current research work, Artificial Fish Swarm algorithm (AFSA) is utilized to train the ANN for automated recognition and classification of stuttering events. The nodes in the ANN layer adapt to the new values based on stochastic and population search to enhance the recognition accuracy and classification, hence the name Adaptive Optimization ANN (AOANN). The major advantage of AOANN includes the ability to recognize and classify stuttered speech compared with other prevailing methods like PSO-ANN, GAANN, and default ANN. Statistical comparison of recognition and classification of stuttered speech was performed by computing MSE, RMSE, and execution time taken by the classifiers. The best result of minimum value of MSE 0.667 and RMSE 0.707107by AOANN was obtained for the average over 100 stuttered speech samples by varying the ratio of training and testing data.
Dysarthria is a motor speech disorder and the most common neurodegenerative disease characterized by low volume in precise articulation, poor coordination of respiratory and pulmonary subsystems, and ...irregular pauses. The key challenge with Dysarthria is reduced intelligibility in speech that forces an individual to undergo numerous socio-professional adversaries due to the lack of self-expression, constrained presentation, limited career opportunities, and accessibility towards the state-of-art advanced technologies, especially voice-controlled activities or human machine interface. Though, numerous efforts have been made to exploit acoustic features from dysarthric speeches to improve intelligibility; however, most of the existing approaches are limited due to change in acoustic features across languages. The approaches suggesting machine learning-driven automatic speech recognition too have failed in retaining most suitable set of acoustic features and learning environment to yield precise word recognition. Considering it as motivation, in this paper, a Hybrid Acoustic Feature-Driven Deep Learning model for Dysarthric Speech Recognition and perceptible speech generation was developed. It exploited Short-Term Fourier Transform driven Mel-Frequency Cepstral Coefficient (MFCC), Pitch, LPC, Gammatone Frequency Cepstral Coefficient (GFCC), Short-Time Energy (STE), Zero-Crossing Rate (ZCR), and short-time ZRC (ST-ZRC) as composite acoustic features to perform LSTM-CNN-based learning for dysarthric speech recognition (DSR) and perceptible speech generation. The statistical performance revealed that the proposed model achieves the highest known accuracy for DSR and ASR systems.
Kannada is an alpha syllabary language wherein each individual akshara/alphabet having different phonetic structures can be divided into sequence of numerous consonants and vowels. In our analysis, ...we concentrate on vowels, fricatives and stop consonants of prerecorded samples of various subjects. The analysis is carried out with respect to basic acoustic features and spectrogram to extract formant frequency and formant analysis to identify the complex natural frequency of the vocal tract functions using linear prediction method.
Dysarthria is a motor speech disorder and the most common neurodegenerative disease characterized by low volume in precise articulation, poor coordination of respiratory and phonatory subsystems, ...irregular pauses and so on. The challenging factor in improving the communication with dysarthria is speech intelligibility. The intelligilibility of speech can be determined by measuring the acoustic characteristics and phonetic structures of speech. In our analysis, we have concentrated on bisyllabic words in the Kannada language. The analysis is carried out with respect to basic acoustic parameters and spectrogram to extract fundamental frequency, formant frequency, jitter, shimmer, HNR and Standard Deviation using PRAAT tool. These results are helpful to identify and differentiate the complex natural frequency of the vocal tract functions with respect to normal subjects and dysarthria subjects. This study provides inputs to increase efficacy in the clinical assessments for monitoring and diagnosing the speech disorders for improving communication among human beings.
PDF
