Speech based region recognition using RNN

Human speech analysis has long been a problem for engineering civilization due to elements including emotional state identification, artificial intelligence study, product assessment, and many more. The two main principles of voice or speech analysis are the identification of an individual's gender and location based on accent. This paper presents feature extraction method from the raw human voice to identify the location that a human voice belongs to. The process comprised a multi-stage pipeline extraction of pertinent features that capture geographical and environmental acoustic signals, including mean, Skewness, standard deviation, and Mel-Frequency Cepstral Coefficients (MFCCs). The primary goal was to identify the phonemes in the Hindi words that were provided. The deep learning model makes use of the large collection of speech samples from the north, south, east, west, and centre areas of India. The retrieved features are used to train the model. The experimental results, which yield 97% classification accuracy using RNN across a variety of regional speech samples, illustrate the efficacy of the proposed design. By sharing our framework and this work, we hope to advance the area of signal processing by inspiring practitioners to experiment with similar systems.