This dataset contains acoustic feature representations extracted from piano recordings in order to analyze the acoustic effects of sustain pedal usage. The dataset is composed of paired audio segments corresponding to pedal-on and pedal-off conditions derived from the same musical performances. A total of 30,257 observations (rows) and 70 feature columns are included in the dataset. Each row represents a time-aligned audio segment extracted from piano recordings. For every acoustic descriptor computed from the pedal condition, the corresponding feature extracted from the no-pedal condition is stored using the same column name with the suffix “.1”. This naming convention allows direct comparison between pedal and non-pedal acoustic characteristics. The dataset contains multiple acoustic descriptors derived from time-domain and spectral-domain analyses commonly used in audio signal processing and music information retrieval. This dataset is intended for research on: piano performance analysis sustain pedal detection acoustic feature comparison machine learning approaches for musical performance analysis Property Value Number of rows 30,257 Number of columns 70 Data type Acoustic feature vectors Format CSV Pedal condition columns original feature names No-pedal condition columns same name + .1 suffix