BAS4R: A Multi-Condition Bangla Speech Dataset for Gender-Aware Real and Spoofed Voice Analysis

BAS4R is a structured and large-scale Bangla speech dataset developed to support research in speech anti-spoofing, replay attack detection, gender-aware voice analysis, and speaker verification robustness evaluation. The dataset contains both authentic real speech and systematically generated spoofed speech recordings collected under controlled and realistic acoustic conditions. The complete corpus comprises approximately 143.88 hours of audio recordings, totaling 120,125 audio files. The dataset is organized into five major categories: Real Data, Channel-based, Effect-based, Replay-based, and Signal Processing. The category-wise distribution includes Real Data with 4,805 files, Channel-based samples with 28,830 files, Effect-based samples with 28,830 files, Replay-based samples with 28,830 files, and Signal Processing samples with 28,830 files. Speech samples were collected from 200 native Bangla speakers, including 110 male and 90 female participants. The speakers were recruited from ten districts of Bangladesh: Barishal, Chapainawabganj, Chittagong, Habiganj, Kishoreganj, Kushtia, Naogaon, Narail, Pabna, and Sylhet. This district-level speaker distribution was designed to capture regional pronunciation, accent variability, and linguistic diversity across Bangla speech. The dataset follows a hierarchical folder structure that separates real and spoofed samples by category, condition, gender, and authenticity label. Spoofed speech samples were generated from real Bangla speech using physical replay setups, real communication channels, effect-based voice modifications, and signal-processing transformations. The spoofing conditions include GSM codec simulation, telephone transmission, low-bitrate VoIP communication, echo, robot, monster, device re-recording, distance replay, speaker-to-microphone replay, pitch shift, formant shift, and time stretch. All audio files are stored in WAV format with a sampling rate of 16 kHz, 16-bit resolution, and mono channel configuration. A centralized metadata file named metadata.csv is also provided. This metadata file maps each audio file path to its corresponding file name, spoofing category, condition, gender, authenticity label, class label, duration, sampling rate, channel configuration, bit depth, and file format. BAS4R is suitable for research in Bangla speech anti-spoofing, replay attack detection, voice authentication security, gender-aware classification, accent-robust spoofing detection, and robust speaker verification under realistic acoustic and channel-degraded conditions.