We present a novel dataset designed to support the development and evaluation of audio restoration techniques focused on historical music recordings. The dataset comprises over 800 audio excerpts, including musicological tape recordings, isolated tape hiss segments, and pitchpipe tones used as tuning references. Each fragment is annotated with metadata such as instrument presence, year of recording, and tape reel number. The dataset enables a variety of restoration-related tasks, including denoising, noise profiling, instrument detection, and segmentation. We also provide baseline results for denoising using state-of-the-art deep learning models and demonstrate an application for playback speed correction using Electrical Network Frequency (ENF) analysis. Our goal is to contribute to the preservation of audio heritage by facilitating reproducible research and benchmarking in music restoration.