遇见数据集

Parametric Overdrive pedal dataset (pOD-set)

收藏
Zenodo2025-09-15 更新2026-05-26 收录
官方服务:

资源简介:

Related Publication: Francesco Ardan Dal Rì, Domenico Stefani, Luca Turchet, Nicola Conci, "Morphdrive: Latent Conditioning For Cross-Circuit Effect Modeling And A Parametric Audio Dataset Of Analog Overdrive Pedals", in Proceedings of the 28-th Int. Conf. on Digital Audio Effects (DAFx25), 2025, Ancona, Italy. Preprint: domenicostefani.com/publications/2025DAFx_Morphdrive_PREPRINT.pdf BibTex: @inproceedings{dalri2025morphdrive, author = "Dal R{\`i}, Francesco A. and Stefani, Domenico and Turchet, Luca and Conci, Nicola", title = "{Morphdrive: Latent Conditioning For Cross-Circuit Effect Modeling And A Parametric Audio Dataset Of Analog Overdrive Pedals}", booktitle = "Proceedings of the 28-th Int. Conf. on Digital Audio Effects (DAFx25)", location = "Ancona, Italy", editor = "Gabrielli, L. and Cecchi, S.", year = "2025", month = "Sept", publisher = "", pages = "23--30", issn = "2413-6689",} Webpage: domenicostefani.com/morphdrive/ Code: github.com/domenicostefani/morphdrive pOD-set is a dataset with 97 hours and 57 minutes of recordings covering 27 overdrive pedals, with 36 combinations of parameters for each. Out of the wide range of different overdrive pedals currently available on the market, we selected a subset of high-end, boutique pedals - spanning from renowned classic circuits to novel, peculiar, and unconventional designs - which in our experience we believe may be representative of the broad spectrum of tonal nuances that characterizes modern guitar. The different parameter settings represent combinations of the gain and tone knobs, typically found in most overdrive pedals. For each knob, we recorded six positions, covering their full range in evenly spaced increments (i.e., 0, 2, 4, 6, 8, and 10 on 10-mark scales). Precise parameter setting was achieved through a custom recording robot equipped with motors (see the related publication here). Additional controls, where present, were set to flat/neutral positions. Detailed settings are provided with the dataset. The dataset was generated by processing a single input audio file through each parameter configuration of each analog pedal. The main input was a ~6-minute WAV file (48 kHz, 24-bit) from Yeh et al. [1], containing a diverse range of instrumental sounds. Moreover, we added three 4-second sinusoidal sweeps (15 Hz to 24 kHz) at different amplitudes (-6, -12, and -24 dBFS), separated by 400 ms of silence to capture the noise floor of each pedal. On each pedal, the volume knob was used to match the output level among different pedals, first manually adjusted to -3dBFS peak at the loudest setting. Since fine-level matching is not feasible due to the circuits nonlinear behavior across different parameters, we adopted per-class normalization of the recordings so that the loudest peak for each pedal was precisely -3 dBFS. Normalization was minimal, with an average amplification of 0.33 ± 0.87 dB. All dataset samples were recorded using a MOTU M4 audio interface. The input signal was routed through a Radial ProRMP passive re-amp box to match the pedals input impedance. We measured and removed the total signal chain latency of 1268 samples from each file for accurate alignment with the input file. Finally, since signal phase is per se tonally irrelevant, we inverted the polarity for phase-inverting circuits to maintain consistency across the dataset. WAV naming convention: Example: a_chime_g3_t5.wav [a]_[chime]_g[3]_t[5].wav A B C D A. Recording type: "a": Main audio from [1], processed through the pedal. "n": Noise floor recording. "s0": Sine sweep (15 Hz to 24 kHz) at -6 dBFS "s1": Sine sweep (15 Hz to 24 kHz) at -12 dBFS "s2": Sine sweep (15 Hz to 24 kHz) at -24 dBFS B. Pedal name. The mapping between this (also called "folder name") and the pedal Model, Brand, and 3-letter code name is provided in the file "0_dataset_metadata.csv" C. Gain knob setting in the range [0-5]. 0 Means knob fully to the left, 5 fully to the right. On a 10-step marked knob, these represent the even values including zero and 10 (0, 2, 4, 6, 8, 10). D. Tone knob setting in the range [0-5]. Pedal List: Index Shortname Brand Model Folder name 1 KOT AnalogMan King of Tone kingoftone 2 HON Bearfoot Honey Bee honeybee 3 BEE Beetronix Overhive overhive 4 WES Bogner Wessex MKII wessex 5 BLU Boss Blues Driver bluesdriver 6 GLA Cornerstone Gladio gladio 7 SS2 Cornish SS2 ss2 8 ELE Dr Scientist The Elements theelements 9 PLU EarthQuaker Devices Plumes plumes 10 OCD Fulltone OCD v1.3 ocd 11 ZEN Hermida Audio Zendrive zendrive 12 PRE Horizon Devices Precision Drive precisiondrive 13 TUB Jam Pedals Tube Dreamer tubedreamer 14 KTR Klon KTR ktr 15 ACQ Lichtlaerm Audio Acquaria aquaria 16 FEL Lunastone Big Fella bigfella 17 CHI Pettyjohn Chime chime 18 LIG Rawkworks Light OD lightod 19 ZOI SviSound Overzoid overzoid 20 DUM Tanabe Dumkudo dumkudo 21 RUM Tone City Big Rumble bigrumble 22 JAN Vemuram Jan Ray janray 23 SIL Vox Silk Drive silkdrive 24 RED Way Huge Red Llama* redllama 25 MOF Wampler Mofetta mofetta 26 385 Walrus Audio 385 MKII 385 27 SOU Xotic Soul Driven souldriven * With tone control mod [1] Y.-T. Yeh, W.-Y. Hsiao, and Y.-H. Yang, “Hyper recurrent neural network: Condition mechanisms for black-box audio effect modeling,” in Proc. 27th Int. Conf. on Digital Audio Effects (DAFx24), Sept. 2024, pp. 97–104 https://www.dafx.de/paper-archive/2024/papers/DAFx24_paper_16.pdf

提供机构:
Zenodo
创建时间:
2025-05-13
二维码
社区交流群
二维码
科研交流群
商业服务