{"ID":2867929,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.18360","arxiv_id":"2509.18360","title":"Speech Vecalign: an Embedding-based Method for Aligning Parallel Speech Documents","abstract":"We present Speech Vecalign, a parallel speech document alignment method that monotonically aligns speech segment embeddings and does not depend on text transcriptions. Compared to the baseline method Global Mining, a variant of speech mining, Speech Vecalign produces longer speech-to-speech alignments. It also demonstrates greater robustness than Local Mining, another speech mining variant, as it produces less noise. We applied Speech Vecalign to 3,000 hours of unlabeled parallel English-German (En-De) speech documents from VoxPopuli, yielding about 1,000 hours of high-quality alignments. We then trained En-De speech-to-speech translation models on the aligned data. Speech Vecalign improves the En-to-De and De-to-En performance over Global Mining by 0.37 and 0.18 ASR-BLEU, respectively. Moreover, our models match or outperform SpeechMatrix model performance, despite using 8 times fewer raw speech documents.","short_abstract":"We present Speech Vecalign, a parallel speech document alignment method that monotonically aligns speech segment embeddings and does not depend on text transcriptions. Compared to the baseline method Global Mining, a variant of speech mining, Speech Vecalign produces longer speech-to-speech alignments. It also demonstr...","url_abs":"https://arxiv.org/abs/2509.18360","url_pdf":"https://arxiv.org/pdf/2509.18360v1","authors":"[\"Chutong Meng\",\"Philipp Koehn\"]","published":"2025-09-22T19:37:20Z","proceeding":"cs.CL","tasks":"[\"cs.CL\"]","methods":"[]","has_code":false}