DarshanaS/IndicAccentDb

1. Introduction

Introducing a novel accent database "IndicAccentDB" which satisfies the below requirements:

• Gender balance: The speech database should be a collection of a wide range of speakers balancing both the male and female speakers to display the characteristics of the speakers speech.

• Phonetically balanced uniform content: To make the classification task simpler and models to distinguish the speakers, we considered building the IndicAccentDB with uniform content, a collection of speech recordings for the Harvard sentences. These sentences gather intrinsic information by combining different phonemes and grammatically focused vocabulary. These sentences are appropriately expressing accents in sentence-level discourse. You can access the Harvard sentences (sample shown below) dataset here: Harvard Sentences recited by the speakers in the recordings.

  The juice of lemons makes fine punch.
  The fish twisted and turned on the bent hook.

• IndicAccentDB contains speech recordings in six non-native English accents of Gujarati, Hindi, Kannada, Malayalam, Tamil, and Telugu. We collected six non-native accents from volunteers who had strong non-native English accents and were well-versed in speaking at least one Indian language. Each speaker was asked to recite the Harvard sentences. The Harvard sentences dataset consists of 72 sets of ten sentences each and is phonetically balanced sentences that are neither too short nor too long.

2. Dataset Usage

To use the dataset in your Python program, refer to the following script:

from datasets import load_dataset
accent_db = load_dataset("DarshanaS/IndicAccentDb")

3. Publications

1. S. Darshana, H. Theivaprakasham, G. Jyothish Lal, B. Premjith, V. Sowmya and K. Soman, "MARS: A Hybrid Deep CNN-based Multi-Accent Recognition System for English Language," 2022 First International Conference on Artificial Intelligence Trends and Pattern Recognition (ICAITPR), Hyderabad, India, 2022, pp. 1-6, doi: 10.1109/ICAITPR51569.2022.9844177.