Dataset Pruning (DP) aims to construct a coreset that achieves performance comparable to the original, full dataset. However, few studies have explored DP in the context of Speech Classification (SC) tasks. Unlike image or text classification, SC is particularly challenging due to the difficulty in capturing the acoustic, semantic, and contextual representations. In this study, we propose a novel dataset pruning method for speech datasets, termed Meltrim, which uses a two-step coarse-to-fine framework designed to address these challenges. Specifically, in Step 1, Meltrim coarsely filters utterance-level redundant samples using DBSCAN clustering on Mel-Frequency Cepstral Coefficients (MFCC) features, which are first flattened and then reduced in dimensionality using UMAP. In Step 2, we perform frame-level redundancy pruning for each utterance via utility pruning, which aims to eliminate irrelevant frames within each utterance. To the best of our knowledge, this is the first dataset pruning approach designed for Speech Classification tasks, demonstrating outstanding performance compared to classical general DP methods. Notably, for the Speech Emotion Recognition, our method achieves up to a 49.5% improvement in WA (Weighted Accuracy) on the MEAD dataset. For the Speaker Identification tasks, it results in a 41.9% reduction in EER (Equal Error Rate) on the VoxCeleb1 dataset.

Pretraining-data attribution typically assigns scalar scores to individual training examples. However, practical data curation involves retaining or removing coherent subsets, making additive pointwise assumptions mathematically restrictive. To address this, we formalize attribution as subset-level counterfactual utility prediction and propose GRASP (Geometry-aware Residual Alignment for Subset Prediction). GRASP combines first-order residual alignment with a low-dimensional train-train geometry term, derived from a smoothness lower bound. By explicitly modeling data interactions via compact 65-dimensional feature sketches, GRASP reduces upfront artifact construction time by nearly an order of magnitude compared to standard baselines. It efficiently scores 100,000 candidate subsets in just 5 seconds on a single CPU, completely bypassing the prohibitive cost of combinatorial retraining. Empirically, GRASP strictly dominates scalable baselines, more than doubling the average task-level subset-utility rank correlation on Linear Datamodeling Score (LDS) benchmarks, while fixed-compute diagnostics on Common Crawl further qualify its practical utility in large-scale curation.