UFTBESD: University of Frontier Technology, Bangladesh-Bangla Emotional Speech Dataset

UFTBESD (University of Frontier Technology, Bangladesh - Bangla Emotional Speech Dataset) is a Bangla language-based speech emotion recognition dataset developed to capture realistic emotional speech in everyday acoustic conditions. The dataset consists of 1400 speech-audio recordings collected from 100 native Bangla speakers aged between 19 and 60 years, with a balanced gender distribution (50 male and 50 female).

Each participant uttered two selected Bangla sentences, with each sentence spoken once (single trial) in seven emotional states: angry, disgust, fear, happy, neutral, sad, and surprise. Thus, the dataset contains 2 sentences × 7 emotions × 100 speakers = 1400 audio clips.

All recordings were collected using smartphone microphones in both indoor and outdoor environments. A significant portion of the data includes natural background noise, making the dataset suitable for real-world speech emotion recognition research. The audio files are stored in WAV format (44.1 kHz, 16-bit, mono). This release focuses on providing raw audio data, and detailed metadata will be added in future versions.

UFTBESD is intended to support the development and evaluation of Bangla speech emotion recognition systems and can be used with common machine learning and deep learning architectures such as CNN, LSTM, BiLSTM, and transformer-based models.

UFTBESD（孟加拉国前沿科技大学孟加拉语情感语音数据集，University of Frontier Technology, Bangladesh - Bangla Emotional Speech Dataset）是一款面向孟加拉语的语音情感识别数据集，旨在捕获日常声学环境下的真实情感语音。该数据集共包含1400条语音音频录音，采集自100名年龄介于19至60岁的孟加拉语母语使用者，性别分布均衡（男性50名，女性50名）。
每位参与者需朗读两段选定的孟加拉语句子，且每段句子分别以愤怒、厌恶、恐惧、开心、中性、悲伤、惊讶七种情感状态各朗读一次（单次试次）。据此，该数据集共包含2段句子×7种情感×100名参与者=1400条音频片段。
所有录音均通过智能手机麦克风在室内与室外环境中采集。该数据集包含大量自然背景噪声，使其适配真实场景下的语音情感识别研究。音频文件以WAV格式存储（采样率44.1kHz，位深度16位，单声道）。本次发布仅提供原始音频数据，详细元数据将在后续版本中补充。
UFTBESD可用于支撑孟加拉语语音情感识别系统的开发与评估，可配合常见的机器学习与深度学习架构使用，例如卷积神经网络（CNN）、长短期记忆网络（LSTM）、双向长短期记忆网络（BiLSTM）以及基于Transformer的模型。