In the digital age, we often face challenges in organizing and searching audio archives of folk music, whose potential frequently remains underutilized due to a lack of metadata and linguistic particularities. In this work, we present a pipeline for automatic indexing and efficient text-based search in extensive archives of folk music field recordings. By using advanced methods of audio segmentation, source separation, speech recognition, and fuzzy string matching, we enable text-based search across recording archives, despite the varied and overlapping types of content, including speech, solo singing, choral singing, and instrumental music. The proposed solution has proven to be an effective tool for searching archives, even with incomplete transcriptions, and represents a significant contribution to the research and preservation of cultural heritage. When evaluated on an artificially degraded corpus, the system achieves 95% accuracy.
|
