Using AI-Based Speech Recognition Systems to Advance Listening and Speaking Skills in French Education

There are now accessible voice recognition systems that use artificial intelligence that can understand French. It takes a lot of effort to compare how an assistant robot can make the most of each. Finding the most accurate French-language voice recognition system in a real-world setting is the primary goal of the project. Several open-source French datasets, commonly used assessment methods, the models utilized by each component of an Automatic Speech Recognition (ASR) system, and how the system operates as a whole are all covered in this study. Using the Word Error Rate (WER) measure, we next compare pre-trained models for French with deep learning-based voice recognition APIs on two separate datasets. In this paper, we proposed new VGG-16 model for speech recognition. The proposed system first learns important features from the signal then recognize the speech with the support of dictionary. For evaluation, we constructed an ASR reference system basing on phoneme modeling utilizing the HMM (Hidden Markov Models) associated with the GMM (Gaussian Mixture Model) models using the HTK tool in order to assess this analysis. The optimal settings for the two primary components of an ASR system—acoustic analysis and decoding—were determined through a battery of tests prior to the system’s installation. We discovered that the optimal parameters leading to acceptable identification accuracy are VGG-16 coefficients with 92.5%, and the trials also suggest that using 25 Gaussian components is a suitable balance between computation time and recognition accuracy.