3 results
Search Results
Now showing 1 - 3 of 3
Item Open Access DEVELOPMENT OF METHODS AND ALGORITHMS TO BUILD A SPEAKER VERIFICATION IN KAZAKH LANGUAGE(СДУ хабаршысы - 2021, 2021) Rashid Sh. ; Kuanyshbay D. ; Nurkey A.Abstract. Speaker verification interfaces are gaining more and more popularity in both academic and commercial industries. It's connected with the latest advances in this area, which can be seen firsthand in our daily life: voice interfaces in computers, robots, cell phones, Internet browsers, and even household appliances. The relevance of the development of Kazakh speech recognition systems arises in connection with the growing needs in the field of public services, provided within the framework of electronic government (egov). Availability voice interfaces will open access to government services for people with disabilities, as well as people living in remote regions and having the only way of access in the form of telephones.Item Open Access SPEECH RECOGNITION BASED ON CONVENTIONAL NEURAL NETWORKS(СДУ хабаршысы - 2021, 2021) Almukhametova D. ; Kuanyshbay D. ; Askar N.Abstract. In this research work, the problem of speech recognition is considered in the form of an analysis of the numbers from 1 to 10 recorded by the speaker on the dictaphone. The paper uses the method of recognizing the spectrogram of an audio signal using convolutional neural networks. Also written and implemented an algorithm for processing input data, and an algorithm for recognizing spoken words. In this work, the quality of recognition was assessed for a different number of convolutional layers. A comparison of the recognition quality is made in cases when the input data for the network are the spectrogram of the audio signal or the first two formants extracted from it. The recognition algorithm was tested using examples of male and female voices with different pronunciation lengths.Item Open Access THE METHODS AND ALGORITHMS FOR RECOGNIZING KAZAKH LANGUAGE FEATURES(СДУ хабаршысы - 2021, 2021) Kalmurzayev Y. ; Kuanyshbay D.; Othman M.Abstract. Despite the importance of automatic speech recognition (ASR), it is difficult to find freely available models, especially for languages with few speakers. This paper describes a method for training Kazakh models based on end-to-end ASR architecture using open-source data. We put the models to the test, and the results are promising. However, much more training data is required to perform well in noisy environments. We make available to the public our trained Kazakh models and training configurations.