Naim Terbeh

Automatic speech correction: A step to speech recognition for people with disabilities

This work consists on achieve an automatic correction system for Arabic continuous speech. This system will be combined by an ASR system for disabled people. For this work, we built a lexicon of 4.000.000 Arabic words through which decides if a word is correct or not. A corpus of Arabic texts is also required to provide a standard summarizing the appearance rate of each two-letter (two-phoneme) in the Arabic language. The results of our system were encouraging and present an advantage to other work for people with articulatory disabilities.

Probabilistic Approach for Detection of Vocal Pathologies in the Arabic Speech

There are different methods for vocal pathology detection. These methods usually have three steps which are feature extraction, feature reduction and speech classification. The first and second steps present obstacles to attain high performance and accuracy of the classification system [20]. Indeed, feature reduction can create a loss of data. In this paper, we present an initial study of Arabic speech classification based on probabilistic approach and distance between reference speeches and speech to classify. The first step in our approach is dedicated to generate a standard distance (phonetic distance) between different healthy speech bases. In the second stage we will determine the distance between speech to classify and reference speeches (phonetic model proper to speaker and a reference phonetic model). Comparing these two distances (distance between speech to classify and reference speeches & standard distance), in the third step, we can classify the input speech to healthy or pathological. The proposed method is able to classify Arabic speeches with an accuracy of 96.25%, and we attain 100% by concatenation falsely classified sequences. Results of our method provide insights that can guide biologists and computer scientists to design high performance systems of vocal pathology detection.

A Novel Approach to Identify Factor Posing Pronunciation Disorders

Literature seems rich with approaches which are based on the features contained in the speech signal and natural language processing techniques to detect vocal pathologies in human speeches. From the literature, we can mention also that several factors (vocal pathology, non-native speaker, psychological state, age …) can pose pronunciation disorders [10]. But to our knowledge, no work has treated pathological speech to identify factor posing pronunciation disorders. The current work consists in introducing an original approach based on the forced alignment score [8] to identify the factor posing mispronunciations contained in the Arabic speech. We distinguish two main factors: the pronunciation disorders can be from native speakers with vocal pathology or from non-native speakers who do not master Arabic-phoneme pronunciation. The results are encouraging; we attain an identification rate of 95 %. Biologists and computer scientists can benefit from our proposed approach to design high performance systems of vocal pathology diagnostic.

Semantic indexing of continuous speech

Speech, as the most important means of communication between human beings, underwent several scientific studies, but until now there remains much to do with this phenomenon. After more than 60 years of research, the representation of speech has not exceeded the representation of the content (standard MPEG7). This paper presents a step towards a standard semantic indexing of continuous speech (indexing by the sense). The index is presented based on the meaning of the elements (Verbs, Names) of the sentences. Also other information that we can extract from speech. The index will be simpler and more meaningful. We tried not to use any external source to represent the meaning, which makes it more difficult and need resources that not in mostly exist. Also it focuses only on the speech, making specific and effective index, and because the speech she is rich enough to have its own standard. This work presents a step towards our goal of a standard semantic indexing continuous speech.

Correction of pathological speeches and assistance to learners with vocal disabilities

This work describes a new methodology for correcting voice defects contained in the Arabic speeches and assisting learners of Arabic vocabulary. For this purpose, we follow four stages. The first step consists in localizing the vocal disabilities which degrade an Arabic voice signal, so we focus on comparing between a referenced probabilistic-phonetic model and a speaker model. Second, we differentiate two cases: Degraded speeches can be generated from pathological problems, or it can be produced by non arabophone learners. Hence, we compare between forced alignment scores. Third, we develop a new algorithm to correct pathological pronunciations. The last task is the conception of an application assisting learners of Arabic vocabulary in improving their pronunciation. The achieved results are encouraging. Moreover, learners of Arabic vocabulary have presented a good amelioration using the developed application. A lot of applications that design systems of voice signal processing can use our proposition.

Identification of Pronunciation Defects in Spoken Arabic Language.

The detection of vocal pathologies is one of the novelties addressing automatic speech processing. There are several intervening approaches that are based on features contained in an acoustic signal and on natural language processing techniques. However, up to our knowledge, these studies are not extended to detect phonemes that pose degraded speeches. In this paper, we propose a new method to detect mispronounced sounds. We are based on a phonetic-probabilistic modeling. The invented study accounts four fundamental tasks. The first task summarizes the calculation of the probabilistic-phonetic model referring to Arabic speech. The second one is dedicated to calculate the probabilistic-phonetic model appropriate to a speaker whose elocution is classified as pathological. Thirdly, we compare between the two previous models to distinguish two main classes: the input speech can be healthy or pathological. The fourth stage consists in introducing an original algorithm based on a phonetic modeling to generate problematic sounds and to evaluate the elocution of each speaker having voice pathologies by attributing them a language level. This task will be only applied if the input speech is pathological. The obtained results are satisfactory. We have attained a problematic-sound identification rate of 96%.

A Robust Algorithm for Pathological-Speech Correction

The current work presents an original approach based on the probabilistic-phonetic modeling to develop an algorithm permitting the correction of pathological Arabic speech. For this purpose, we follow three steps. The first consists in detecting the voice defects manifesting in the Arabic speech. Second, the sounds begetting degraded speeches are identified. The last task consists in proposing an original algorithm based on probabilistic-phonetic modeling to correct the pathological pronunciations. The developed algorithm is highly efficient. Indeed, we have attained a correction performance of 97%. Accordingly, researchers in computer sciences, in speech therapy and in biology can support in our contribution to the pathological speeches processing.

Arabic speech analysis to identify factors posing pronunciation disorders and to assist learners with vocal disabilities

The literature seems rich with studies addressing the detection of pronunciation disorders. The features contained in the speech signal and natural language processing techniques present famous parameters used for this objective. Despite the diversity of factors posing pronunciation disorders (vocal pathologies, non-native speakers, psychological state, age, etc.), no work has been extended to identify these factors and to assist speakers with pronunciation defects in learning spoken languages. The current work presents an original approach based on the probabilistic-phonetic modeling of Arabic speech to detect vocal disorders [1]. If the analyzed speech presents some degradations, the forced alignment score technique will be introduced to distinguish between two main factors that pose mispronunciations. Pronunciation defects can be from a native speaker suffering from vocal pathology or from a non-native speaker who learns the spoken Arabic language as an L2. Also, a platform is developed to assist speakers with degraded speeches in learning the spoken Arabic language. The present work accounts five steps. The first step consists in calculating the referenced phonetic model of the Arabic speech. This model will be used in detecting the vocal defects contained in the Arabic speech. Second, the referenced forced alignment scores for Arabic phonemes are calculated. In the third phase, for each new speaker with vocal disorders, their forced alignment scores of non-problematic phonemes are calculated [10]. In the fourth step, the two previous scores are compared to distinguish between the pronunciation disorders caused by native speakers suffering from vocal pathologies and by non-native speakers who do not master Arabic-phoneme pronunciation. The last phase consists in developing a platform to assist speakers with pronunciation defects to learn the spoken Arabic language. We are satisfied with the obtained results. We have attained an identification rate of factors posing pronunciation disorders of 95%, and the speakers using our platform have shown a good progression. Speech therapists, biologists and computer scientists can benefit from this work to develop performant systems of pathological speech processing: pathological speech recognition, accent evaluation, e-learning, etc.

Probabilistic Approach to Arabic Speech Correction for Peoples with Language Disabilities

This work consists in achieving an automatic speech correction system for continuous Arabic speech with large vocabulary in mono-speaker mode. Two vectors to be generated: the first is an Arabic speech standard (probability of occurrence of each Arabic bi-phoneme), the second gives a probabilistic representation of the speech sequence to be corrected. Using these two vectors, phonemes that pose pronunciation problems to speakers and their replacements can be identified. The rest is a game of substitutions and belonging tests to an Arabic lexicon. For that, an acoustic model for Arabic speech and a lexicon of 4 million distinct words have been built. Results of the work were encouraging and present a reference for other works for people with language disabilities. A correction rate of 97% is reached. Probabilistic Approach to Arabic Speech Correction for Peoples with Language Disabilities