SGPA

SGPA (Spectrogram-Guided Phonetic Alignment) aligns an audio signal to phonetic units so that audio segments, rather than individual frames, become players; it makes Shapley explanations of speech input feasible. alignment implements the alignment (SpectrogramGuidedAligner); the partitioner that makes segments players is planned.

Source: arXiv 2603.02250 (listed in the legacy MLLM-Shap README). Roadmap: Roadmap.