{"paper":{"title":"Asymmetric Encoder-Decoder Based on Time-Frequency Correlation for Speech Separation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SR-CorrNet separates speech by splitting coarse separation into the encoder and progressive reconstruction into a shared-weight decoder that interacts across speakers.","cross_cats":["cs.SD"],"primary_cat":"eess.AS","authors_text":"Hyung-Min Park, Ui-Hyeop Shin","submitted_at":"2026-03-31T00:37:15Z","abstract_excerpt":"Speech separation in realistic acoustic environments remains challenging because overlapping speakers, background noise, and reverberation must be resolved simultaneously. Although recent time-frequency (TF) domain models have shown strong performance, most still rely on late-split architectures, where speaker disentanglement is deferred to the final stage, creating an information bottleneck and weakening discriminability under adverse conditions. To address this issue, we propose SR-CorrNet, an asymmetric encoder-decoder framework that introduces the separation-reconstruction (SepRe) strategy"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results on WSJ0-{2,3,4,5}Mix, WHAMR!, and LibriCSS demonstrate consistent improvements across anechoic, noisy-reverberant, and real-recorded conditions in both single- and multi-channel settings, highlighting the effectiveness of TF-domain SepRe with correlation-based filter estimation for speech separation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the proposed asymmetric encoder-decoder with SepRe strategy and cross-speaker interaction in the decoder will reliably avoid information bottlenecks and yield better speaker discriminability than late-split architectures under adverse conditions, without the gains depending on dataset-specific tuning or post-hoc architectural choices.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SR-CorrNet introduces an asymmetric TF-domain architecture with separation-reconstruction strategy and correlation-to-filter estimation that yields consistent gains on WSJ0-Mix, WHAMR!, and LibriCSS under anechoic, noisy-reverberant, and real-recorded conditions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SR-CorrNet separates speech by splitting coarse separation into the encoder and progressive reconstruction into a shared-weight decoder that interacts across speakers.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"edf3a4d68f78da21e5d33975980b7107558cd774809cb2e0947fd54374d8aed6"},"source":{"id":"2603.29097","kind":"arxiv","version":2},"verdict":{"id":"85cfdbdc-aa2d-49f6-b763-d9bc3ec42054","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:34:25.998525Z","strongest_claim":"Experimental results on WSJ0-{2,3,4,5}Mix, WHAMR!, and LibriCSS demonstrate consistent improvements across anechoic, noisy-reverberant, and real-recorded conditions in both single- and multi-channel settings, highlighting the effectiveness of TF-domain SepRe with correlation-based filter estimation for speech separation.","one_line_summary":"SR-CorrNet introduces an asymmetric TF-domain architecture with separation-reconstruction strategy and correlation-to-filter estimation that yields consistent gains on WSJ0-Mix, WHAMR!, and LibriCSS under anechoic, noisy-reverberant, and real-recorded conditions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the proposed asymmetric encoder-decoder with SepRe strategy and cross-speaker interaction in the decoder will reliably avoid information bottlenecks and yield better speaker discriminability than late-split architectures under adverse conditions, without the gains depending on dataset-specific tuning or post-hoc architectural choices.","pith_extraction_headline":"SR-CorrNet separates speech by splitting coarse separation into the encoder and progressive reconstruction into a shared-weight decoder that interacts across speakers."},"references":{"count":79,"sample":[{"doi":"","year":2018,"title":"TasNet: time-domain audio separation network for real-time, single-channel speech separation,","work_id":"92026762-7c3d-4b91-8f94-2cc0eaa86251","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Conv-TasNet: Surpassing ideal time–frequency magnitude mask- ing for speech separation,","work_id":"2c10b83f-aa19-4a3c-8b4e-09553a379c35","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Dual-Path RNN: Efficient long sequence modeling for time-domain single-channel speech separation,","work_id":"8174ce5e-4aba-4966-b659-384472cd849f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Attention is all you need in speech separation,","work_id":"b9ae55b7-b39c-4c6b-b4da-783227775cbd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"TFPSNet: Time-frequency domain path scanning network for speech separation,","work_id":"25d4eb93-0d65-47ae-b8da-73d80cdc5a47","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":79,"snapshot_sha256":"f97fc82a667b1b9eea78ad1096c5549e2e24f26c579d0b3026ec3a50665a02f0","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}