{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:BMQNALFCOUX6AIDEBQF2KY6BUJ","short_pith_number":"pith:BMQNALFC","canonical_record":{"source":{"id":"2606.21458","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-19T14:14:20Z","cross_cats_sorted":[],"title_canon_sha256":"bb17795a6318614fcc20f54e6202c35d2e8b1faccd90228b958b91b68b71789b","abstract_canon_sha256":"fe7c7d497a9caac3a91b925684b15a1c99e40ed7744d933fbf72e2d8fb65fc4f"},"schema_version":"1.0"},"canonical_sha256":"0b20d02ca2752fe020640c0ba563c1a246edbc30202e74a3653212ba390f1884","source":{"kind":"arxiv","id":"2606.21458","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.21458","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"arxiv_version","alias_value":"2606.21458v1","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.21458","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_12","alias_value":"BMQNALFCOUX6","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_16","alias_value":"BMQNALFCOUX6AIDE","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_8","alias_value":"BMQNALFC","created_at":"2026-06-23T01:13:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:BMQNALFCOUX6AIDEBQF2KY6BUJ","target":"record","payload":{"canonical_record":{"source":{"id":"2606.21458","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-19T14:14:20Z","cross_cats_sorted":[],"title_canon_sha256":"bb17795a6318614fcc20f54e6202c35d2e8b1faccd90228b958b91b68b71789b","abstract_canon_sha256":"fe7c7d497a9caac3a91b925684b15a1c99e40ed7744d933fbf72e2d8fb65fc4f"},"schema_version":"1.0"},"canonical_sha256":"0b20d02ca2752fe020640c0ba563c1a246edbc30202e74a3653212ba390f1884","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:13:11.001393Z","signature_b64":"1gK1gMNvkA+vjdMqnx4nMpQIySpQBKwP7Kt8QCFH1+T3mg9eXftjzGVoimQiWi4y6ijwQOxT/O/+yYr+fUK3Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0b20d02ca2752fe020640c0ba563c1a246edbc30202e74a3653212ba390f1884","last_reissued_at":"2026-06-23T01:13:11.000936Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:13:11.000936Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2606.21458","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:13:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"G1HWchmxPoV+NQ8PHE43f91Fat7u+Z9CztfyK1QXVnR0vx3XC3QzihXvnnSf94qmBSk1DUF/ZvZ1Y6XzAPmxBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-04T10:57:51.360706Z"},"content_sha256":"c792c5c3327f7daaf6f8e3c9cebf37746ae61a37491e5d776e9ad8c5fe9469a4","schema_version":"1.0","event_id":"sha256:c792c5c3327f7daaf6f8e3c9cebf37746ae61a37491e5d776e9ad8c5fe9469a4"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:BMQNALFCOUX6AIDEBQF2KY6BUJ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Post-Training Speech Enhancement Language Models with Perceptual Rewards","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Antonis Asonitis, Fr\\'ed\\'eric Berdoz, Luca A. Lanzend\\\"orfer, Roger Wattenhofer","submitted_at":"2026-06-19T14:14:20Z","abstract_excerpt":"Speech enhancement language models achieve strong results when trained on discrete audio tokens, but their optimization relies on token-level cross-entropy rather than the perceptual metrics used for evaluation. We introduce a post-training stage for autoregressive speech enhancement language models using Group Sequence Policy Optimization (GSPO) with multi-metric perceptual rewards. Our method directly optimizes non-differentiable quality metrics (DNSMOS, WER, and UTMOS) as reward signals, without learned surrogates or offline preference pairs. Applied to two autoregressive base models, UniSE"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.21458","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2606.21458/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:13:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3ZeA5bT0Ykq8aWeeGF434lKPdVQWkqQgMHdWoM3Mt4Fo1Wb2PEElVi/hJ85cmdDbsKfUuseLUdd/rvtPJ//vBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-04T10:57:51.361074Z"},"content_sha256":"530b9ddc6afe74b84eea7e99e31fc8bbed5fac6313ac9d5b8f92c901687472dc","schema_version":"1.0","event_id":"sha256:530b9ddc6afe74b84eea7e99e31fc8bbed5fac6313ac9d5b8f92c901687472dc"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/bundle.json","state_url":"https://pith.science/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-04T10:57:51Z","links":{"resolver":"https://pith.science/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ","bundle":"https://pith.science/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/bundle.json","state":"https://pith.science/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/BMQNALFCOUX6AIDEBQF2KY6BUJ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:BMQNALFCOUX6AIDEBQF2KY6BUJ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"fe7c7d497a9caac3a91b925684b15a1c99e40ed7744d933fbf72e2d8fb65fc4f","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-19T14:14:20Z","title_canon_sha256":"bb17795a6318614fcc20f54e6202c35d2e8b1faccd90228b958b91b68b71789b"},"schema_version":"1.0","source":{"id":"2606.21458","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.21458","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"arxiv_version","alias_value":"2606.21458v1","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.21458","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_12","alias_value":"BMQNALFCOUX6","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_16","alias_value":"BMQNALFCOUX6AIDE","created_at":"2026-06-23T01:13:11Z"},{"alias_kind":"pith_short_8","alias_value":"BMQNALFC","created_at":"2026-06-23T01:13:11Z"}],"graph_snapshots":[{"event_id":"sha256:530b9ddc6afe74b84eea7e99e31fc8bbed5fac6313ac9d5b8f92c901687472dc","target":"graph","created_at":"2026-06-23T01:13:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2606.21458/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Speech enhancement language models achieve strong results when trained on discrete audio tokens, but their optimization relies on token-level cross-entropy rather than the perceptual metrics used for evaluation. We introduce a post-training stage for autoregressive speech enhancement language models using Group Sequence Policy Optimization (GSPO) with multi-metric perceptual rewards. Our method directly optimizes non-differentiable quality metrics (DNSMOS, WER, and UTMOS) as reward signals, without learned surrogates or offline preference pairs. Applied to two autoregressive base models, UniSE","authors_text":"Antonis Asonitis, Fr\\'ed\\'eric Berdoz, Luca A. Lanzend\\\"orfer, Roger Wattenhofer","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-19T14:14:20Z","title":"Post-Training Speech Enhancement Language Models with Perceptual Rewards"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.21458","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c792c5c3327f7daaf6f8e3c9cebf37746ae61a37491e5d776e9ad8c5fe9469a4","target":"record","created_at":"2026-06-23T01:13:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"fe7c7d497a9caac3a91b925684b15a1c99e40ed7744d933fbf72e2d8fb65fc4f","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-19T14:14:20Z","title_canon_sha256":"bb17795a6318614fcc20f54e6202c35d2e8b1faccd90228b958b91b68b71789b"},"schema_version":"1.0","source":{"id":"2606.21458","kind":"arxiv","version":1}},"canonical_sha256":"0b20d02ca2752fe020640c0ba563c1a246edbc30202e74a3653212ba390f1884","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0b20d02ca2752fe020640c0ba563c1a246edbc30202e74a3653212ba390f1884","first_computed_at":"2026-06-23T01:13:11.000936Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-23T01:13:11.000936Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"1gK1gMNvkA+vjdMqnx4nMpQIySpQBKwP7Kt8QCFH1+T3mg9eXftjzGVoimQiWi4y6ijwQOxT/O/+yYr+fUK3Dw==","signature_status":"signed_v1","signed_at":"2026-06-23T01:13:11.001393Z","signed_message":"canonical_sha256_bytes"},"source_id":"2606.21458","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c792c5c3327f7daaf6f8e3c9cebf37746ae61a37491e5d776e9ad8c5fe9469a4","sha256:530b9ddc6afe74b84eea7e99e31fc8bbed5fac6313ac9d5b8f92c901687472dc"],"state_sha256":"ab7db84b6f4fb4892a31d046a68cf304f99f97d39b168e7ac9f507e1cb81c187"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6ZM4HwnV/mGMIF4X0GbkHuusHMwc6AfazW+1q8oMSp9OrFItM7DuG8viNiK8CsIMu+0zkCdVRipeKkqp4R23DA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-04T10:57:51.362904Z","bundle_sha256":"e99535956e33f1abbd7b1ba07a307027b63638d99ea750dd0b96006f70b88f25"}}