BlueData: A Dataset of Perceptual Audio Degradations Caused by Wireless Packet Loss

BlueData is a dataset of clean and degraded music segments recorded under controlled wireless Bluetooth transmission conditions. Audio degradation results from real packet loss induced by signal attenuation. The dataset consists of 2-second audio clips, each labeled as: 0 → Clean (no audible degradation)1 → Degraded (perceptual artifacts caused by Bluetooth packet loss) All labels were assigned through human auditory inspection by certified QA testers.This dataset supports research on: Perceptual sound event detection Anomaly detection in audio Wireless audio quality assessment in black-box testing environments Annotation Process Each audio segment was labeled through a two-pass auditory inspection process. Initially, one QA tester with ISTQB CTFL certification listened to each segment and labeled it as either clean or degraded. A second certified tester independently reviewed all labels. In cases of disagreement, the final label was determined through discussion and consensus. This process ensured perceptual validity and consistency across the dataset. Label encoding:0 = Clean (no audible degradation)1 = Degraded (perceptual artifact present due to Bluetooth packet loss) Dataset Coverage & Structure Training Set: train/0/ → 11,760 clean audio segments train/1/ → 3,757 degraded audio segments Test Set: test/ → 3,880 mixed segments Original Audio Properties Source: Free Music Archive (FMA) Format: WAV, mono, 44.1 kHz Segment duration: 2 seconds Bitrate: Varies (typically 192–320 kbps MP3 before conversion) Tracks were selected from a variety of musical genres and converted to WAV format under uniform preprocessing conditions.This variation reflects the diversity found in real-world streaming content. Notes No other types of distortion (e.g., clipping, filtering, compression) were introduced. Only real artifacts from Bluetooth transmission loss. The genre distribution was not balanced or controlled, as this dataset is intended for perceptual degradation detection rather than genre classification. Acknowledgments This dataset was developed through a collaboration between the Federal University of Amazonas (UFAM), the Center for Advanced Systems of Recife (CESAR), and MOTOROLA. It was produced with support from projects developed in the Manaus Free Trade Zone. According to SUFRAMA regulations, the author acknowledges this support and states compliance with Lei Federal nº 8.387/1991. Citation Citation If you use this dataset in your work, please cite both the dataset and the associated peer-reviewed publication: Dataset: Guimarães, Victória de Souza. BlueData: A Dataset of Perceptual Audio Degradations Caused by Wireless Packet Loss (Version 1.0). Zenodo, 2025. https://doi.org/10.5281/zenodo.15801604 Associated publication: Guimarães, Victória; Bentes, Luiz; Pires, Ana; de Freitas, Rosiane. Perceptual Detection of Packet Loss-Induced Audio Artifacts in Black-Box Wireless Music Systems. In: Proceedings of the 10th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE 2025), Barcelona, Spain, 2025, p. 215–219. https://doi.org/10.5281/zenodo.17251589 BibTeX @dataset{guimaraes2025bluedata, author = {Guimar{\~a}es, Vict{\'o}ria de Souza}, title = {BlueData: A Dataset of Perceptual Audio Degradations Caused by Wireless Packet Loss}, year = {2025}, version = {1.0}, publisher = {Zenodo}, doi = {10.5281/zenodo.15801604}, url = {https://doi.org/10.5281/zenodo.15801604}} @inproceedings{guimaraes2025dcase, author = {Guimar{\~a}es, Vict{\'o}ria and Bentes, Luiz and Pires, Ana and de Freitas, Rosiane}, title = {Perceptual Detection of Packet Loss-Induced Audio Artifacts in Black-Box Wireless Music Systems}, booktitle = {Proceedings of the 10th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE 2025)}, address = {Barcelona, Spain}, pages = {215--219}, year = {2025}, doi = {10.5281/zenodo.17251589}, url = {https://doi.org/10.5281/zenodo.17251589}, isbn = {978-84-09-77652-8}}