| Autor: |
G, Thimmaraja Yadava, G, Nagaraja B, S, Jayanna H |
| Zdroj: |
Multimedia Tools & Applications; Jan2024, Vol. 83 Issue 2, p4195-4217, 23p |
| Abstrakt: |
This research work showcases advancements in an isolated Kannada automatic speech recognition (ASR) system designed for accessing agricultural commodity prices and weather information in uncontrolled environments. The system includes an interactive voice response system (IVRS), models of ASR, and databases of weather and agricultural commodity prices information. However, the previous system suffered from reduced accuracy due to the presence of various background noises during offline and online speech recognition. To address this issue, the proposed system includes a background noise reduction module that is introduced before the part of speech feature extraction. The investigation results indicate that the proposed noise reduction algorithm outperforms traditional signal processing algorithms, resulting in no audibility of musical and other background noises in the enhanced NOIZEUS speech corpora and isolated Kannada speech data. The use of this noise suppression algorithm and time delay neural network (TDNN) ASR modeling technique in the system results in a 1.1% improvement in speech recognition accuracy compared to the previous deep neural network - hidden Markov model (DNN-HMM) based system. The enhanced isolated Kannada system was tested online by 500 speakers/users for accessing real-time agricultural commodity prices and weather information in Kannada language/dialects under corrupted environments. The algorithms source code and ASR models are made publicly available. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink