Spoken Language Identification Using ConvNets

Abstract

Language Identification (LI) is an important first step in several speech processing systems. With a growing number of voice-based assistants, speech LI has emerged as a widely researched field. To approach the problem of identifying languages, we can either adopt an implicit approach where only the speech for a language is present or an explicit one where text is available with its corresponding transcript. This paper focuses on an implicit approach due to the absence of transcriptive data. This paper benchmarks existing models and proposes a new attention based model for language identification which uses log-Mel spectrogram images as input. We also present the effectiveness of raw waveforms as features to neural network models for LI tasks. For training and evaluation of models, we classified six languages (English, French, German, Spanish, Russian and Italian) with an accuracy of 95.4% and four languages (English, French, German, Spanish) with an accuracy of 96.3% obtained from the VoxForge dataset. This approach can further be scaled to incorporate more languages.

Spoken Language Identification Using Deep Learning

“…The attributes of the proposed method are represented in Table 6 . The trial and error method is used while running the convolution neural network [ 8 , 14 ], word embedding Keras [ 34 , 35 ], and Naïve Bayes [ 36 – 38 ]. The selection of hyperparameter is also defined as an NP-complete problem [ 39 , 40 ].…”

“…Various state-of-the-art results on various audio classification tasks have been obtained by using log-Mel spectrograms of raw audio, like features, which convert the audio utterance into images [ 8 ]. CNN gives an excellent performance gain in classification on these features [ 14 ]. The motivation of work has come from these studies.…”

“…The process of spoken language identification using the CNN technique uses spectrograms of raw audio signals as input to a convolutional neural network (CNN) [ 8 , 14 ]. A spoken language identification dataset is collected and preprocessed for the training phase.…”

Low-Resource Spoken Language Identification Using Self-Attentive Pooling and Deep 1D Time-Channel Separable Convolutions

“…In a more traditional linguistic setting, Sarthak et al [34] explore 1D-ConvNet that auto-extracts and classifies features from raw audio input and 2D-ConvNet architectures, and enhance the performance of these approaches by utilizing Mixup augmentation of inputs and attention mechanism. They achieve 93.7% and 95.4% overall accuracy on a six language (En, Fr, De, Es, Ru, It) dataset with overlapping phonemes based on the VoxForge [38] dataset.…”

Spoken language identification using i-vectors, x-vectors, PLDA and logistic regression

“…Sarthak, et al [21] used raw audio waveforms as sound input which provide increased performance by avoiding overheads in calculating log-Mel spectrum for each audio file. This study uses the convolutional neural networks classification method because based on the journals studied, convolutional neural networks have very good performance compared to other machine learning techniques, the sound input data used is raw audio waveform because it is quite popular, because raw audio waveforms have advantages.…”