Fusion of multi-stream speech features for dialect classification

Sinha, Shweta; Jain, Aruna; Agrawal, S.

doi:10.1007/s40012-015-0063-y

Cited by 7 publications

(1 citation statement)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This study focuses on examining the impact acoustic characteristics of vowel on four distinct Hindi dialects. By examining formants (F1, F2 and F3), pitch (F0), and pitch slope data, the researchers explore the eight vowel's acoustic characteristics from Hindi language [46]. Support Vector Machine (SVM) models have proven to be highly effective for prediction and classification tasks, especially when dealing with high-dimensional input spaces.…”

Section: Literature Reviewmentioning

confidence: 99%

Assamese Dialect Identification Using Static and Dynamic Features from Vowel

Das,

Bhattacharjee

2024

JAIT

View full text Add to dashboard Cite

This paper introduces a novel method for identifying Assamese dialects by analyzing the acoustic and prosodic aspects of vowel sounds in speech signals. The distinctive characteristics of these dialects are captured through the use of acoustic parameters such as formants (F1, F2, and F3), as well as prosodic features like energy, fundamental frequency (F0), and duration. To evaluate this approach, a comprehensive vowel speech corpus is collected from native Assamese speakers representing four different dialectal regions. Frame-level statistical features are extracted from vowel sounds, while temporal dynamic features are obtained from steady-state vowel segments. The data collection process involves using a phonetically rich script to record both read and spontaneous speech interactions from speakers of the four dialects. Various classification methods, including three decision tree-based classifiers, i.e., Random Forest (RF), Extreme Random Forest (ERF), and Extreme Gradient Boosting (XGB), are applied to distinguish the four dialects. The performance of each feature, whether static or dynamic, is individually evaluated. The study reveals that the identification of Assamese dialects is influenced by factors such as speech length, intensity, pitch, and formant frequencies. To assess the significance of these features in distinguishing dialects and to measure their combined impact on the identification system, single-factor Analysis of Variance (ANOVA) tests are conducted. Notably, when static features are combined with the Extreme Random Forest (ERF) ensemble model, the overall accuracy of dialect identification reaches 77%. This research demonstrates the efficacy of using acoustic and prosodic features to accurately classify Assamese dialects, shedding light on the subtle variations within them. In summary, this paper provides a robust framework for Assamese dialect identification and contributes to our understanding of dialect discrimination, paving the way for more advanced dialect identification systems.

show abstract

Section: Literature Reviewmentioning

confidence: 99%