Automatic Spoken Language Recognition with Neural Networks

Abstract

Abstract-Translation has become very important in our generation as people with completely different cultures and languages are networked together through the Internet. Nowadays one can easily communicate with anyone in the world with the services of Google Translate and/or other translation applications. Humans can already recognize languages that they have priory been exposed to. Even though they might not be able to translate, they can have a good idea of what the spoken language is. This paper demonstrates how different Neural Network models can be trained to recognize different languages such as French, English, Spanish, and German. For the training dataset voice samples were choosed from Shtooka, VoxForge, and Youtube. For testing purposes, not only data from these websites, but also personally recorded voices were used. At the end, this research provides the accuracy and confidence level of multiple Neural Network architectures, Support Vector Machine and Hidden Markov Model, with the Hidden Markov Model yielding the best results reaching almost 70 percent accuracy for all languages.

Hybrid Feature Selection Method Based on Harmony Search and Naked Mole-Rat Algorithms for Spoken Language Identification From Audio Signals

“…In this paper, the work is presented in three main subsections namely Pre-processing (IV-A), Feature Extraction Cimarusti et al [12] Polynomial classification using LPC features 8 84% Jerry T. Foil [10] Formant and prosodic feature-based language identification 3 64% Marc A. Zissman [13] HMM based language identification 20 92% M. Sugiyama [14] Vector Quantization Technique based language identification 20 80% Gazeau et al [15] HMM based language identification 4 70% Revay et al [16] log-Mel spectra based DNN approach for language identification 6 89% Bartz et al [18] CRNN based language identification 6 91% Mukherjee et al [20] MFCC-2 based language identification 3 98.09%…”

“…Nowadays, in an attempt to move beyond low-level spectral analysis, several attempts have been made for better and meaningful feature extraction techniques that rely heavily upon deep learning models for language identification purposes. Gazeau et al [15], in their study, used Neural Network (NN), Support Vector Machine (SVM) and HMM models to recognize 4 different languages namely French, English, Spanish and German. The dataset was prepared by using voice samples from Shtooka, VoxForge and Youtube.…”

“…For all the three databases, RF classifier is found to outperform the remaining four classifiers, therefore, in the rest of the section, the obtained results is presented using RF classifier only. The parameter setting for these 10 FS methods is illustrated in Table (15) From the results obtained from Table ( 16), it is observed that for CSS10 database, HS-NMR based FS method achieves the best classification accuracy of 99.89% while BCS achieves the least accuracy of 98.11%. For the remaining 2 databases also, the method outperforms the remaining 10 algorithms achieving accuracies of 98.22% and 99.75% for VoxForge and Indic TTS databases respectively.…”

Is Attention always needed? A Case Study on Language Identification from Speech

“…Researcher [Gazeau and Varol 2018] established the use of Neural Network, Support Vector Machine, and Hidden Markov Model (HMM) to identify different languages. Hidden Markov models converts speech into a sequence of vectors and was used to capture temporal features in speech.…”

Flattening Layer Pruning in Convolutional Neural Networks

“…Calculations made with their help can be found, among others, in forecasting air pollution [5] or when conversing via cell phone [6]. A popular and recent development is the use of neural networks in natural language processing, in particular, for text generation [7], automatic text translation [8], text analysis [9], spam message detection [10] and spoken text recording [11]. Due to their versatility and the possibility of modeling non-linear processes, ANNs are used in the automotive industry (navigation systems, autopilot), telecommunications and robotics [12].…”