What question did this study set out to answer?

The aim is to improve Hindi speech signals contaminated by various background noises using deep learning techniques.

Speech Enhancement using Fully Connected Deep Neural Network for Hindi Speech Corrupted by Nonstationary Noises

View Full Paper

Ask AI

Bookmark

Authors

Demonstrates enhancements in speech quality in Hindi using deep learning techniques under noisy conditions, highlighting effectiveness.

The aim is to improve Hindi speech signals contaminated by various background noises using deep learning techniques.
Utilized a fully connected deep neural network for speech enhancement.
Created a diverse Hindi speech database including stationary and nonstationary noises.
Added background noises at varying input Signal-to-Noise Ratio levels to simulate real-world conditions.
Evaluated improvements in SNR, PESQ, and STOI metrics to assess system effectiveness.
Achieved a significant SNR improvement of 13.94 db in highly nonstationary machine gun noise at -5 db SNR level.
Obtained a PESQ score of 2.91, indicating better speech quality.
STOI score reached 0.94, demonstrating high intelligibility retention across noise types.

Gupta et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark