摘要
ODAQ 是一个旨在解决公开可获取的音频信号及其对应主观质量评分数据集稀缺问题的数据集。该数据集包含 240 个音频样本，以及通过 MUSHRA 听觉测试获得的相应质量评分。这些经过质量评分的音频样本均为原始音频素材的处理版本（原始音频素材亦已公开提供）。原始音频素材包括：采样频率为 44.1 kHz 或 48 kHz 的立体声音频；共 14 段音乐片段（其中 8 段为独奏录音）；11 段类似电影配乐的音频片段，包含对白、音乐与音效的混合（同时提供独立音轨及对应的文本转录）。本文中，该数据集用于客观音频质量评估方法的基准测试。
基准测试
| 基准 | 方法 | 指标 |
|---|---|---|
| audio-quality-assessment-on-odaq-open-dataset | NMR | Pearson correlation coefficient (PCC): 0.89 |
| audio-quality-assessment-on-odaq-open-dataset | PEAQ (ODG) | Pearson correlation coefficient (PCC): 0.87 |
| audio-quality-assessment-on-odaq-open-dataset | SMAQ | Pearson correlation coefficient (PCC): 0.77 |
| audio-quality-assessment-on-odaq-open-dataset | ViSQOLAudioV3 | Pearson correlation coefficient (PCC): 0.77 |
| audio-quality-assessment-on-odaq-open-dataset | DNSMOS (OVRL) | Pearson correlation coefficient (PCC): 0.38 |
| audio-quality-assessment-on-odaq-open-dataset | SI-SDR | Pearson correlation coefficient (PCC): 0.44 |
| audio-quality-assessment-on-odaq-open-dataset | PEAQ-CSM | Pearson correlation coefficient (PCC): 0.89 |
| audio-quality-assessment-on-odaq-open-dataset | 2f-model | Pearson correlation coefficient (PCC): 0.87 |
| audio-quality-assessment-on-odaq-open-dataset | PESQ | Pearson correlation coefficient (PCC): 0.74 |