Abstract:Identifying the language spoken in an audio source is the difficult task of automatic language identification (LID) in speech processing. Short audio segments pose a significant challenge in language identification because they contain limited contextual information and fewer distinguishing features compared to longer audio samples. This lack of context makes it difficult to accurately identify the language, as the model has less data to analyse. By addressing the challenge of short-duration audio, the research aims to develop more robust and versatile language identification systems that can operate effectively even with minimal input. Another objective of the research is to address the specific challenge of identifying Indian languages accurately and efficiently from short-duration audio segments using CNNs and spectrogram representations in Python. The methodology involves several key steps: initially, audio data undergoes pre-processing to normalize the signals and reduce noise, ensuring consistency across the dataset. Subsequently, the audio signals are converted into spectrograms, which offer a visual depiction of the frequency spectrum, capturing both temporal and frequency characteristics essential for language discrimination. A CNN model is then built and trained using these spectrograms, with a specific architecture designed to extract significant features from the spectrograms. The system's performance is evaluated on a custom dataset consisting of three Indian languages: Hindi, Tamil, and Malayalam. The experimental findings show that a 98.9% accuracy rate is attained by the CNN-based model, surpassing the performance of existing models. The proposed method has potential applications in areas such as automatic speech recognition and speaker identification, where accurate and efficient language identification is crucial.

Acoustics Based Intent Recognition Using Discovered Phonetic Units for Low Resource Languages

Intent Recognition and Unsupervised Slot Identification for Low Resourced Spoken Dialog Systems

Intent Classification Using Pre-trained Language Agnostic Embeddings For Low Resource Languages

Leveraging Acoustic and Linguistic Embeddings from Pretrained speech and language Models for Intent Classification

Low-Resource Language Identification From Speech Using Transfer Learning

Improvement of Acoustic Models Fused with Lip Visual Information for Low-Resource Speech

Real-time Caller Intent Detection In Human-Human Customer Support Spoken Conversations

Accent Recognition with Hybrid Phonetic Features

Exploiting Spectral Augmentation for Code-Switched Spoken Language Identification

Streaming Language Identification using Combination of Acoustic Representations and ASR Hypotheses

Listen with Intent: Improving Speech Recognition with Audio-to-Intent Front-End

Phone-aware Neural Language Identification

Multilingual and Cross-Lingual Intent Detection from Spoken Data

Creating Spoken Dialog Systems in Ultra-Low Resourced Settings

Acoustic Modeling Based on Deep Learning for Low-Resource Speech Recognition: An Overview

Multi-Lingual Attention based Multi-Intent Detection in Dialogue System

ELAICHI: Enhancing Low-resource TTS by Addressing Infrequent and Low-frequency Character Bigrams

Convolutional neural network based language identification system: A spectrogram based approach

Investigating the Impact of Cross-lingual Acoustic-Phonetic Similarities on Multilingual Speech Recognition

Towards Building ASR Systems for the Next Billion Users