Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech

Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech Naim Terbeh, Mounir Zrigui LaTICE Laboratory Monastir-Tunisia naim.terbeh@gmail.com, mounir.zrigui@fsm.rnu.tn Abstract We propose in this work a novel acoustic phonetic study for Arabic people suffering from language disabilities and non-native learners of Arabic language to classify Arabic continuous speech to pathological or healthy and to identify phonemes that pose pronunciation problems (case of pathological speeches). The main idea can be summarized in comparing between the phonetic model reference to Arabic spoken language and that proper to concerned speaker. For this task, we use techniques of automatic speech processing like forced alignment and artificial neural network (ANN) (Basheer, 2000). Based on a test corpus containing 100 speech sequences, recorded by different speakers (healthy/pathological speeches and native/foreign speakers), we attain 97% as classification rate. Algorithms used in identifying phonemes that pose pronunciation problems show high efficiency: we attain an identification rate of 100%. Keywords: Arabic healthy/pathological speech, language disabilities, phonetic model, forced alignment, artificial neural network, pronunciation problems 1. Introduction Analysis of characteristics generated from speech signal produced by a speaker may be used to classify this speech to healthy or pathological. Our project focuses in introducing a new probabilistic approach that aims to detect vocal pathologies in the Arabic speech and identify phonemes that pose pronunciation problems. Nowadays, speech therapists use different medical techniques in vocal pathologies detection. Laryngoscopy, Electromyography and Videokimography are the most used (Majidnesha, 2013). But these methods possess a number of disadvantages; in application of these diagnostic methods, the patients feel much discomfort which can distort the produced signal so that it may lead to incorrect diagnosis (Alenso, 2001; Jollife, 2008). Acoustic and phonetic modeling seems the most appropriate and efficient in this area. We use the phonetic transcription to generate a phonetic model of Arabic speech: the percentage of occurrence of each bi-phoneme in Arabic spoken language. The comparison between the reference phonetic model (numerical model (Zouaghi, 2008)) of Arabic speech and that specific to the concerned speaker lead to classify the speech produced by this latter to healthy or pathological. 2. Stat of the Art In the literature, there are several studies that treat human speeches to detect pronunciation disorders. Also, there are several approaches which are based on features contained  in the speech signal: Vahid and al. in (Majidnesha, 2013), propose an ANN based approach to classify speeches to healthy or pathological. The proposed method accounts three stages which are extraction of MFCC (Ihichaichareon, 2012) coefficients vector, using the PCA method (Ihichaichareon, 2012; Jollife, 2008) to reduce feature vector (MFCC vector) and use an ANN to classify speeches in input (healthy or pathological).  Little and al. in (little, 2006) combine between linear classification and biophysics of speech production to online vocal pathologies detection.  In the (Majidnesha, 2012) work, Vahid proposes a HMM-based approach to classify speech to healthy or pathological. This method accounts three steps which are extraction of MFCC vector, use the LBG algorithm (patane, 2001) to extract the …

Content truncated to 500 words