Isolated Bengali Vowel and Word Speech Sounds

Published: 13 September 2021| Version 1 | DOI: 10.17632/2h6975kdsx.1


This dataset includes two separate datasets---one for vowel and another for words. Seven isolated Bengali vowels (/অ/ [/ɔ/], /আ/ [/a/], /ই/ [/i/], /উ/ [/u/], /ঋ/ [/ri/], /এ/ [/e/], /ঐ/ [/oi/]) and seven isolated Bengali words (বোতল, বন, কপি, দোকোন, শেষ, সঠিক, উপরে) were collected at Khulna University of Engineering and Technology, Khulna 9203, Bangladesh. We collected these speech sounds from twenty male and female volunteers (age 20-26 years) having Bengali as their first language. We guided the speakers to pronounce each vowel and word in a different local accent along with a normal accent. Thus, we produced 40 data in each class of the two datasets, i.e., each of the vowel classes (অ, আ, etc.) contains 40 data and each of the word class (বোতল, বন, etc.) also contains 40 data.


Steps to reproduce

The first step was to capture the vowels as a continuous stream with silence between two successive vowels. Since we recorded sounds on a typical smartphone (Xiaomi Redmi 3), the raw captured sounds were two-channel, and each file contained all the vowels as a single audio file for each speaker. In order to further process the raw sounds, we used Audacity software (version 2.2.2), where at first, we converted the recorded stereo channel audio to the mono channel by using the functionality available at "Tracks > Stereo Track to Mono." Furthermore, we clipped the individual vowels and words in the same software by observing the waveform and hearing it simultaneously. Finally, we exported and saved the selected audio as 32-bit float datatype in different classes. The above steps explain the process of creating the vowel dataset. We followed the exact process for creating the word dataset.


Khulna University of Engineering and Technology


Speech Analysis, Bengali Language