Spoken Language Identification and Translation using Deep Learning

Identifying the language spoken in an audio clip by a speaker using a model is practiced here as Spoken language identification. Here the tedious task is to identify the important features that differentiate languages. The proposed system collects numerous audios and converts to images of spectrogram using convolutional neural network algorithms to detect the language. The main objective of the proposed system is to detect the languages spoken by the speaker along with the gender identification of the speaker. Experiments will be applied on diverse audio archives dataset “spoken language identification” from Kaggle. Here, the audio files are encompassed with utterances for 10 seconds. The dataset is divided into training set and testing sets. The proposed system also identifies the gender of the speaker and translation of the identified language into English. It consists of three convolutions followed by a fully-connected layer. The first two convolutions are for feature extraction, and the last one is for classification. RNN is applied to classify the gender of the speaker in an audio recording and Seq2Seq model is trained to translate the identified language into English.