A Multimodal Method for Detecting Language through Speech in Ten Indian Languages

This research presents the development of an audio-driven language prediction system aimed at recognizing and classifying Indian languages. The Random Forest machine learning model is employed by the system. It is developed using a unique dataset that includes audio samples from ten Indian languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Punjabi, Tamil, Telugu, and Urdu. To improve the accuracy of language identification, the audio data is subjected to preprocessing steps, such as segmentation and noise reduction. Key audio features, including Mel-frequency cepstral coefficients (MFCCs), zero-crossing rate, spectral centroid, and chroma attributes, are obtained to represent the acoustic qualities of the languages. These features are then utilized to train the Random Forest classifier, which shows excellent effectiveness in distinguishing between the different languages. The suggested system was evaluated through accuracy metrics, demonstrating significant capability in correctly categorizing audio inputs according to their respective languages. This study tackles the challenges of multilingual speech recognition in a linguistically diverse country such as India and offers a scalable method for automatic language identification.