{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JYNIXK3XQQCH2USUYHQENAVCZQ","short_pith_number":"pith:JYNIXK3X","schema_version":"1.0","canonical_sha256":"4e1a8bab7784047d5254c1e04682a2cc30ea6419e47b1aab353d80f6b9011107","source":{"kind":"arxiv","id":"2604.12383","version":2},"attestation_state":"computed","paper":{"title":"On the Distillation Loss Functions of Speech VAE for Unified Reconstruction, Understanding, and Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Joint-marginal alignment with adaptive weighting delivers the best overall performance in speech VAEs for reconstruction, understanding, and generation.","cross_cats":[],"primary_cat":"cs.SD","authors_text":"Changhao Cheng, Dongya Jia, Jian Wu, Wangyou Zhang, Wei Wang, Yanmin Qian, Zhuo Chen","submitted_at":"2026-04-14T07:17:55Z","abstract_excerpt":"Continuous speech representations based on Variational Autoencoders (VAEs) have emerged as a promising alternative to traditional spectrogram or discrete token based features for speech generation and reconstruction. Recent research has tried to enrich the structural information in VAE latent representations by aligning with self-supervised learning (SSL) features, aiming for better generation performance. However, it remains unclear whether the widely-used alignment approach based on time-axis distillation is optimal when considering more tasks. To address this problem, this paper systematica"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2604.12383","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.SD","submitted_at":"2026-04-14T07:17:55Z","cross_cats_sorted":[],"title_canon_sha256":"82349ea94e3e70bf653e41213946caaebcc3b025b80fe039aa3df472bc4762ac","abstract_canon_sha256":"c86a63430c97848d704c52a9362623eeb92ed65049648e5fe725e04a312f0952"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:10.692042Z","signature_b64":"5yHXvdVUfiLa4YKzxHADEHydmrM61LA085K8sX0jXYfQ7vKJIeZrN8OlwNY1g1MXokJKvlzZVkJ00V3LK6muCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e1a8bab7784047d5254c1e04682a2cc30ea6419e47b1aab353d80f6b9011107","last_reissued_at":"2026-05-26T02:04:10.691278Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:10.691278Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Distillation Loss Functions of Speech VAE for Unified Reconstruction, Understanding, and Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Joint-marginal alignment with adaptive weighting delivers the best overall performance in speech VAEs for reconstruction, understanding, and generation.","cross_cats":[],"primary_cat":"cs.SD","authors_text":"Changhao Cheng, Dongya Jia, Jian Wu, Wangyou Zhang, Wei Wang, Yanmin Qian, Zhuo Chen","submitted_at":"2026-04-14T07:17:55Z","abstract_excerpt":"Continuous speech representations based on Variational Autoencoders (VAEs) have emerged as a promising alternative to traditional spectrogram or discrete token based features for speech generation and reconstruction. Recent research has tried to enrich the structural information in VAE latent representations by aligning with self-supervised learning (SSL) features, aiming for better generation performance. However, it remains unclear whether the widely-used alignment approach based on time-axis distillation is optimal when considering more tasks. To address this problem, this paper systematica"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Extensive experiments show that the joint-marginal alignment approach with adaptive weighting can achieve the best overall performance while allowing for a controllable balance.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen SSL features and evaluation metrics adequately represent the full space of reconstruction, understanding, and generation needs without hidden task-specific biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Joint-marginal alignment plus adaptive weighting in speech VAE distillation yields the best combined performance on reconstruction, understanding, and generation tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Joint-marginal alignment with adaptive weighting delivers the best overall performance in speech VAEs for reconstruction, understanding, and generation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0b3a074b4933d879fcd648a71cded6ff2beb8353b2db73708ae6bc7a493e1509"},"source":{"id":"2604.12383","kind":"arxiv","version":2},"verdict":{"id":"6051e709-55c8-4aa7-a170-11814bb4bd1e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:27:09.623362Z","strongest_claim":"Extensive experiments show that the joint-marginal alignment approach with adaptive weighting can achieve the best overall performance while allowing for a controllable balance.","one_line_summary":"Joint-marginal alignment plus adaptive weighting in speech VAE distillation yields the best combined performance on reconstruction, understanding, and generation tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen SSL features and evaluation metrics adequately represent the full space of reconstruction, understanding, and generation needs without hidden task-specific biases.","pith_extraction_headline":"Joint-marginal alignment with adaptive weighting delivers the best overall performance in speech VAEs for reconstruction, understanding, and generation."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.12383/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2604.12383","created_at":"2026-05-26T02:04:10.691413+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.12383v2","created_at":"2026-05-26T02:04:10.691413+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.12383","created_at":"2026-05-26T02:04:10.691413+00:00"},{"alias_kind":"pith_short_12","alias_value":"JYNIXK3XQQCH","created_at":"2026-05-26T02:04:10.691413+00:00"},{"alias_kind":"pith_short_16","alias_value":"JYNIXK3XQQCH2USU","created_at":"2026-05-26T02:04:10.691413+00:00"},{"alias_kind":"pith_short_8","alias_value":"JYNIXK3X","created_at":"2026-05-26T02:04:10.691413+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.16964","citing_title":"SemaVoice: Semantic-Aware Continuous Autoregressive Speech Synthesis","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12383","citing_title":"On the Distillation Loss Functions of Speech VAE for Unified Reconstruction, Understanding, and Generation","ref_index":2,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ","json":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ.json","graph_json":"https://pith.science/api/pith-number/JYNIXK3XQQCH2USUYHQENAVCZQ/graph.json","events_json":"https://pith.science/api/pith-number/JYNIXK3XQQCH2USUYHQENAVCZQ/events.json","paper":"https://pith.science/paper/JYNIXK3X"},"agent_actions":{"view_html":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ","download_json":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ.json","view_paper":"https://pith.science/paper/JYNIXK3X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.12383&json=true","fetch_graph":"https://pith.science/api/pith-number/JYNIXK3XQQCH2USUYHQENAVCZQ/graph.json","fetch_events":"https://pith.science/api/pith-number/JYNIXK3XQQCH2USUYHQENAVCZQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ/action/storage_attestation","attest_author":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ/action/author_attestation","sign_citation":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ/action/citation_signature","submit_replication":"https://pith.science/pith/JYNIXK3XQQCH2USUYHQENAVCZQ/action/replication_record"}},"created_at":"2026-05-26T02:04:10.691413+00:00","updated_at":"2026-05-26T02:04:10.691413+00:00"}