Bangla Isolated Speech Dataset

In this work, a dataset is created containing 36 Bangla words and 24 English words with the help of 25 different persons from different regions of Bangladesh. 30 samples per word from both male and female speakers are recorded. The number of speech samples in the created dataset is 1800. The samples are recorded using a sound recorder (smartphones) in a room environment. Among them 1200 samples are used for the training dataset and 600 samples are used for test dataset. All the training samples are recorded as “.wav" files and the test samples are recorded as ".mp3" files. The speech sample words are recorded with a high sampling frequency (44.10 kHz) to create this dataset.

Follwing words used to create the dataset. The folders are arranged as per the same order of following words. The words in this dataset are [Bangla word (pronunciation, meaning)]:
34 Bangla words (folder no s0-s35) (pronunciation: meaning):
দাঁড়াও (Darao: Stand), হাটো (Hato: Walk), থামো (Thamo: Stop), সামনে (Shamne: Forward), পেছনে (Pechone: Backward), ডানে (Dane: Right), বামে (Bame: Left), উপরে (upore: Up), নিচে (niche: Down), শুরু (shuru: Start), শেষ (Shesh: End), পড় (Poro: Read), লিখ (Likho: Write), শুনো (Shuno: Listen), বলো (Bolo: Speak), তাকাও (Takao: Look), আলো (Alo: Light), শব্দ (Shobdo: Sound), সময় (Shomoy: Time), ভর (Vor: Mass), নাম (Nam: Name), বই (Boi: Book), খাতা (Khata: Pad), কলম (Kolom: Pen), গাড়ি (Gari: Car), বাড়ি (Bari: House), পানি (Pani: Water), খাবার (Khabar: Food), কুকুর (Kukur: Dog), বিড়াল (Biral: Cat), মানুষ (Manush: Human), শিশু (Shishu: Child), হাতুড়ি (Haturi: Hammer), চশমা (Shoshma: Glasses), গান (Gan: Song), ঘড়ি (Ghori: Watch),
24 English words (folders s36-s59):
WiFi, Class, Program, Diode, Capacitor, Switch, Television, Radio, Light, Mobile, Head-phone, Mike, Battery, Internet, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9.

本研究构建了一个双语语音数据集，共收录36个孟加拉语单词与24个英语单词。该数据集由来自孟加拉国不同地区的25名受试者参与录制。针对每个单词，分别录制了来自男性与女性发音者的共30条语音样本，数据集总语音样本量达1800条。所有样本均在室内环境下通过智能手机录音设备录制。其中1200条样本被划分为训练集，剩余600条样本作为测试集；训练集样本以".wav"格式存储，测试集样本则采用".mp3"格式存储。为保障语音数据质量，所有样本均采用44.10 kHz的高采样率进行录制。

本数据集所用单词按「孟加拉语单词（发音，含义）」的格式标注，文件夹的命名顺序与下述单词的排列顺序完全一致。数据集收录的单词如下：
1. 34个孟加拉语单词（对应文件夹s0-s35）：
দাঁড়াও（Darao：站立）、হাটো（Hato：行走）、থামো（Thamo：停止）、সামনে（Shamne：向前）、পেছনে（Pechone：向后）、ডানে（Dane：向右）、বামে（Bame：向左）、উপরে（upore：向上）、নিচে（niche：向下）、শুরু（shuru：开始）、শেষ（Shesh：结束）、পড়（Poro：阅读）、লিখ（Likho：书写）、শুনো（Shuno：聆听）、বলো（Bolo：说话）、তাকাও（Takao：看向）、আলো（Alo：光线）、শব্দ（Shobdo：声音）、সময়（Shomoy：时间）、ভর（Vor：质量）、নাম（Nam：名称）、বই（Boi：书籍）、খাতা（Khata：笔记本）、কলম（Kolom：钢笔）、গাড়ি（Gari：汽车）、বাড়ি（Bari：房屋）、পানি（Pani：水）、খাবার（Khabar：食物）、কুকুর（Kukur：狗）、বিড়াল（Biral：猫）、মানুষ（Manush：人类）、শিশু（Shishu：儿童）、হাতুড়ি（Haturi：锤子）、চশমা（Shoshma：眼镜）、গান（Gan：歌曲）、ঘড়ি（Ghori：钟表）
2. 24个英语单词（对应文件夹s36-s59）：WiFi、Class、Program、Diode、Capacitor、Switch、Television、Radio、Light、Mobile、Head-phone、Mike、Battery、Internet、0、1、2、3、4、5、6、7、8、9