Identification of Indian languages using multi-level spectral and prosodic features

Reddy, V. Ramu; Maity, Sudhamay; Rao, K. Sreenivasa

doi:10.1007/s10772-013-9198-0

Cited by 71 publications

(6 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The mel scale, which is shown in Figure 12, compares the perceived frequency, or pitch, of a pure tone to its measured frequency (Reddy et al, 2013). Humans are much better at detecting slight pitch fluctuations at low frequencies than they are at high frequencies.…”

Section: Mel Scalementioning

confidence: 99%

Longitudinal Investigation of Work Stressors Using Human Voice Features

Natarajan¹,

Maheswaran²,

Dhanalakshmi³

et al. 2023

AIP

View full text Add to dashboard Cite

Stress is a part of everyone's life. Any event or thought that makes you upset, furious or anxious can set it off. It will affect the human health mentally and physically and produce a negative impact on nervous and immune systems in our body. The human voice carries a lot of information about the person speaking. It also aids in determining a person's current state. In this proposed method, stress was detected using a deep learning model. Automatic stress detection is becoming an intriguing study topic as the necessity for communication between humans and intelligent systems rises. The hormone called cortisol can also be used to determine the body's stress state. For most people, however, it is not a viable option. Speech features are particularly affected by stress, which is combined with the aim that voice data would serve as an easy-to-capture measure of everyday human stress levels and hence as an early warning signal of stress-related health problems. The proposed technique extracts Mel filter bank spectral coefficients from pre-processed voice input and the spectrum coefficients are extracted. The features of Mel frequency cepstral coefficients are applied to feed-forward networks and long short-term memory to predict the status of stress output using a binary decision, i.e., unstressed or stressed. The Mel spectrum and spectrogram output shows the variation in stressed and unstressed voice features. The results of the proposed method indicate better performance compared to an existing model. The model was developed as a web application to be used by workers to test their state of stress at any time.

show abstract

Section: Mel Scalementioning

confidence: 99%

Longitudinal Investigation of Work Stressors Using Human Voice Features

Natarajan¹,

Maheswaran²,

Dhanalakshmi³

et al. 2023

AIP

View full text Add to dashboard Cite

show abstract

“…This is known as weighted score level fusion. Adaptive weighted combination scheme [32] has been used for combining the scores of speech mode classification (SMC) models. In this work, weights are derived by seeking the range of weights for each model present in a respective stage (see Figure 4) on the development set and the final optimal weights are used for evaluation.…”

Section: Score Level Fusionmentioning

confidence: 99%

Multilingual and multimode phone recognition system for Indian languages

Tripathi

Reddy

Rao

2020

Speech Communication

View full text Add to dashboard Cite

The aim of this paper is to develop a flexible framework capable of automatically recognizing phonetic units present in a speech utterance of any language spoken in any mode. In this study, we considered two modes of speech: conversation, and read modes in four Indian languages, namely, Telugu, Kannada, Odia, and Bengali. The proposed approach consists of two stages: (1) Automatic speech mode classification (SMC) and(2) Automatic phonetic recognition using mode-specific multilingual phone recognition system (MPRS). In this work, vocal tract and excitation souce features are considered for speech mode classification (SMC) task. SMC systems are developed using multilayer perceptron (MLP). Further, vocal tract, excitation source, and tandem features are used to build the deep neural network (DNN)-based MPRSs. The performance of the proposed approach is compared with mode-dependent MPRSs. Experimental results show that the proposed approach which combines both SMC and MPRS into a single system outperforms the baseline mode-dependent MPRSs.

show abstract

“…As a consequence, training state-of-theart DNN-based LID systems using such small corpus is prone to overfitting. To address this issue, researchers have experimented with different approaches, such as applying data augmentation, feature selection, regularizers, cross-validation, or model ensembling [17,70,78]. Still, achieving generalization for the LID systems with low-resource languages remains the key challenge [17].…”

Section: Introductionmentioning

confidence: 99%