Automatic Language Identification from Non-Uniform Region Using Bi-LSTM and CNN

Language Identification is the way to identify the language of a speech or text sample. It entails dissecting the traits of the text or speech and contrasting them with recognized patterns and models of various languages. Language identification is crucial for a variety of processes, including text-to-voice synthesis, natural language processing, and speech recognition. The basic goal of this project is to accurately identify the language from the speech sample. The three Indian languages of Malayalam, Assamese, and Hindi are represented by a language identification system that uses features from non-uniform speech regions to represent phonotactic differences. The BiLSTM and DCNN classification methods are used to identify the given speech with the relevant attributes. The suggested technique starts by extracting spectral information from the speech sample. In addition, we extract prosodic characteristics and spectral features and categorize them using BiLSTM and DCNN. The research, which focuses on northeastern languages, makes use of the recorded databases for three Indian languages. The bi-LSTM model has an accuracy score of 82 % higher than the DCNN model.