Level-specific auditory-model targets (90 dB) generated from TIMIT TRAIN speech using the Verhulst et al. (2018) preprocessing path in CoNNear_IHC-ANF

# Dataset OverviewThis record describes a derived dataset of level-specific auditory-model targets generated from TIMIT TRAIN speech using the preprocessing path based on the Verhulst et al. (2018) reference model as included in the CoNNear_IHC-ANF framework. -ENG This record does not present a new auditory model. Instead, it provides standardized derived outputs produced by batch-processing speech utterances through an existing auditory-model-based preprocessing pipeline and exporting the resulting targets in NumPy format. # Source FrameworkThe data were generated using the repository:HearingTechnology/CoNNear_IHC-ANF Relevant references:- Drakopoulos, F., Baby, D., & Verhulst, S. (2021). A convolutional neural-network framework for modelling auditory sensory cells and synapses. Communications Biology, 4, 827. DOI: 10.1038/s42003-021-02341-5- Verhulst, S., Altoè, A., & Vasilkov, V. (2018). Computational modeling of the human auditory periphery: Auditory-nerve responses, evoked potentials and hearing loss. Hearing Research, 360, 55-75. DOI: 10.1016/j.heares.2017.12.018 # Processing Procedure1. Speech waveforms were loaded from the TIMIT TRAIN directory and grouped by dialect region (DR) subset.2. Each waveform was processed using the function `get_tl_target`.3. A single level-specific slice was selected from the returned outputs (index 9 in the local implementation, corresponding in this project setup to the 90 dB condition).4. Two outputs were retained: - transmission-line target (`tl`) - stimulus target (`stim`)5. Both outputs were standardized to a fixed temporal length of 2560 samples.6. If a sequence was shorter than 2560 samples, edge padding was applied.7. If a sequence was longer than 2560 samples, it was truncated.8. Outputs were saved separately for each DR subset in NumPy `.npy` format. # Data ContentsFor each processed DR subset, the dataset contains:- `tl_train_DR{n}_90db.npy`: shape `(N, 2560, 201)`- `stim_train_DR{n}_90db.npy`: shape `(N, 2560)` where `N` is the number of utterances processed in the corresponding DR subset. # Intended UseThis dataset is intended for:- non-commercial academic research,- reproducible preprocessing in auditory-model experiments,- educational use in speech and hearing modeling workflows. # Important Licensing and Access NoteThis record describes data derived from:1. model code distributed under the UGent Academic License, and2. speech material originating from the TIMIT corpus. Because redistribution rights may be constrained by the upstream licenses and agreements, the files associated with this record are shared in restricted form. Access should only be granted for non-commercial academic research purposes and, where applicable, to users with legitimate access rights to the original source materials. # DisclaimerThis dataset should not be interpreted as a new auditory model or an independently developed cochlear simulation framework. It is a derived, standardized output set generated from an existing preprocessing pipeline. # 数据集概述本记录描述了一个衍生数据集,该数据集是使用基于 Verhulst 等人 (2018) 参考模型的预处理路径,从 TIMIT TRAIN 语音生成的特定级别听觉模型目标,该参考模型包含在 CoNNear_IHC-ANF 框架中。-中文 本记录并未提出新的听觉模型。相反,它提供了通过现有的基于听觉模型的预处理流程对语音进行批量处理而生成的标准化衍生输出,并将结果目标导出为 NumPy 格式。 # 数据源框架数据由以下存储库生成:HearingTechnology/CoNNear_IHC-ANF 相关参考文献:- Drakopoulos, F., Baby, D., & Verhulst, S. (2021). A convolutional neural-network framework for modelling auditory sensory cells and synapses. Communications Biology, 4, 827. DOI: 10.1038/s42003-021-02341-5 - Verhulst, S., Altoè, A., & Vasilkov, V. (2018). Computational modeling of the human auditory periphery: Auditory-nerve responses, evoked potentials and hearing loss. Hearing Research, 360, 55-75. DOI: 10.1016/j.heares.2017.12.018 # 处理流程1. 从 TIMIT TRAIN 目录加载语音波形,并按方言区 (DR) 子集分组。2. 使用 `get_tl_target` 函数处理每个波形。3. 从返回的输出中选择一个特定电平的切片(本地实现中的索引 9,在本项目设置中对应于 90 dB 条件)。4. 保留两个输出: - 传输线目标 (`tl`) - 刺激目标 (`stim`) 5. 将两个输出标准化为 2560 个样本的固定时间长度。6. 如果序列长度小于 2560 个样本,则应用边缘填充。7. 如果序列长度大于 2560 个样本,则将其截断。8. 将每个 DR 子集的输出分别以 NumPy `.npy` 格式保存。 # 数据内容对于每个已处理的 DR 子集,数据集包含:- `tl_train_DR{n}_90db.npy`:形状为 `(N, 2560, 201)` - `stim_train_DR{n}_90db.npy`:形状为 `(N, 2560)` 其中“N”是相应DR子集中处理的语句数量。 # 预期用途此数据集旨在用于:- 非商业性学术研究,- 听觉模型实验中的可重复预处理,- 语音和听觉建模工作流程中的教育用途。 # 重要许可和访问说明本记录描述了源自以下数据的数据:1. 根据 UGent 学术许可证分发的模型代码,以及2. 源自 TIMIT 语料库的语音材料。 由于上游许可和协议可能限制了再分发权限,因此与此记录相关的文件以受限形式共享。访问权限仅限于非商业性学术研究用途,并且在适用情况下,应授予拥有原始资料合法访问权限的用户。 # 免责声明:本数据集不应被解释为新的听觉模型或独立开发的耳蜗模拟框架。它是由现有预处理流程生成的衍生标准化输出数据集。