{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:KCXG5LP5BUZSBAOZEMVTN3Z5PU","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":"cd2790a0db990c662e3544ea8ad77ba01427a8f6d561f01a89a1d7d1454cfb20","cross_cats_sorted":["cs.CL","cs.SD"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"eess.AS","submitted_at":"2026-01-30T10:12:16Z","title_canon_sha256":"539e452adb9e8cad4111a9bbe956f431d5145aca67b5f57c3d0557c80e80ca06"},"schema_version":"1.0","source":{"id":"2601.22792","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.22792","created_at":"2026-05-18T02:45:05Z"},{"alias_kind":"arxiv_version","alias_value":"2601.22792v2","created_at":"2026-05-18T02:45:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.22792","created_at":"2026-05-18T02:45:05Z"},{"alias_kind":"pith_short_12","alias_value":"KCXG5LP5BUZS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"KCXG5LP5BUZSBAOZ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"KCXG5LP5","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:fdf9cd8cf4e030cd7ff3712880593c6c5c582f4e2659a8a81fe43a016fd4d66a","target":"graph","created_at":"2026-05-18T02:45:05Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"On two-speaker mixtures, CALM reduces biased word error rate (B-WER) from 12.7 to 4.7 on LibriSpeech2Mix and biased character error rate (B-CER) from 16.6 to 8.4 on CSJMix2 (eval3), demonstrating the effectiveness of joint acoustic-linguistic modeling across languages."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the simulated two-speaker mixtures (LibriSpeechMix, CSJMix) and the IHM-mix condition of AMI sufficiently represent the acoustic and linguistic statistics of real overlapping conversations where speaker turns, noise, and context vary more widely."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"CALM jointly models acoustic speaker identity and linguistic context to cut biased error rates by more than half on two-speaker English and Japanese mixtures."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"CALM integrates speaker embeddings for target extraction with dynamic vocabulary biasing to halve biased error rates in overlapping multi-speaker ASR."}],"snapshot_sha256":"a2e62ec2c927effb30f1b17bca70f7d9f2104e1d5fdba533710054e6a7c08a08"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"372247d71b14197b1b40475030f8ccb99c8b2aeb05830ab6a298b23308cf8939"},"paper":{"abstract_excerpt":"We present CALM, a joint Contextual Acoustic-Linguistic Modeling framework for multi-speaker automatic speech recognition (ASR). In personalized AI scenarios, the joint availability of acoustic and linguistic cues naturally motivates the integration of target-speaker conditioning with contextual biasing in overlapping conversations. CALM implements this integration in an end-to-end framework through speaker embedding-driven target-speaker extraction and dynamic vocabulary-based contextual biasing. We evaluate CALM on simulated English (LibriSpeechMix) and Japanese (Corpus of Spontaneous Japane","authors_text":"Chikara Maeda, Chyi-Jiunn Lin, Muhammad Shakeel, Shinji Watanabe, Yosuke Fukumoto","cross_cats":["cs.CL","cs.SD"],"headline":"CALM integrates speaker embeddings for target extraction with dynamic vocabulary biasing to halve biased error rates in overlapping multi-speaker ASR.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"eess.AS","submitted_at":"2026-01-30T10:12:16Z","title":"CALM: Joint Contextual Acoustic-Linguistic Modeling for Personalization of Multi-Speaker ASR"},"references":{"count":53,"internal_anchors":1,"resolved_work":53,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"INTRODUCTION Single-speaker automatic speech recognition (ASR) systems have achieved state-of-the-art (SOTA) performance across many speech- processing tasks [1–3]. However, in multi-speaker settings ","work_id":"dcdd3d30-4ad3-4dd1-9847-c32484f4b15c","year":null},{"cited_arxiv_id":"2601.22792","doi":"","is_internal_anchor":true,"ref_index":2,"title":"CALM: Joint Contextual Acoustic-Linguistic Modeling for Personalization of Multi-Speaker ASR","work_id":"76604f8e-13fd-429c-9922-df6583dfac5b","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Frame-level target- speaker activity posteriors are computed as: P vad =σ(W vad ˆH(L) +b vad),(11) withP vad ∈[0,1] T enc","work_id":"1b015532-8761-49ec-9892-37edae9d41a3","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"EXPERIMENTS The CALM framework is built on ESPnet [45], pairing a Conformer encoder with a Transformer decoder. The Conformer has 12 lay- ers with 4 heads and 1024 linear units (kernel size 31) and ap","work_id":"1d55f36d-0f28-4a57-b4eb-7773cdc825ce","year":2048},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"However, unlike in simulated conditions, overall WER in- creases from 37.4 to 39.1 absolute points. Our error analysis indi- cates that this degradation is primarily driven by an increase in inser- ti","work_id":"020dd479-0292-4a34-b2b0-c20ff1f0891f","year":null}],"snapshot_sha256":"44546e885cd008a81f565798e10d28d5b34db6c030c1b4dfc303290113d923a6"},"source":{"id":"2601.22792","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-16T09:47:23.362902Z","id":"830856c6-8e34-4e42-86b6-e7286d86de38","model_set":{"reader":"grok-4.3"},"one_line_summary":"CALM jointly models acoustic speaker identity and linguistic context to cut biased error rates by more than half on two-speaker English and Japanese mixtures.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"CALM integrates speaker embeddings for target extraction with dynamic vocabulary biasing to halve biased error rates in overlapping multi-speaker ASR.","strongest_claim":"On two-speaker mixtures, CALM reduces biased word error rate (B-WER) from 12.7 to 4.7 on LibriSpeech2Mix and biased character error rate (B-CER) from 16.6 to 8.4 on CSJMix2 (eval3), demonstrating the effectiveness of joint acoustic-linguistic modeling across languages.","weakest_assumption":"That the simulated two-speaker mixtures (LibriSpeechMix, CSJMix) and the IHM-mix condition of AMI sufficiently represent the acoustic and linguistic statistics of real overlapping conversations where speaker turns, noise, and context vary more widely."}},"verdict_id":"830856c6-8e34-4e42-86b6-e7286d86de38"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:258683f0f5f566345d08a5112953414e3caadbc96dced08d31bcfb2f45aa0383","target":"record","created_at":"2026-05-18T02:45:05Z","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":"cd2790a0db990c662e3544ea8ad77ba01427a8f6d561f01a89a1d7d1454cfb20","cross_cats_sorted":["cs.CL","cs.SD"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"eess.AS","submitted_at":"2026-01-30T10:12:16Z","title_canon_sha256":"539e452adb9e8cad4111a9bbe956f431d5145aca67b5f57c3d0557c80e80ca06"},"schema_version":"1.0","source":{"id":"2601.22792","kind":"arxiv","version":2}},"canonical_sha256":"50ae6eadfd0d332081d9232b36ef3d7d15a57379dc8f8b4303a5fe05bf752bb9","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"50ae6eadfd0d332081d9232b36ef3d7d15a57379dc8f8b4303a5fe05bf752bb9","first_computed_at":"2026-05-18T02:45:05.698741Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:45:05.698741Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HrzWkoFy2QLJsP+i4ml4NpKFpb9LR1foekdoTKqp7oHxXII4843xrS1gOSe8Q9VursPBum4GrbYc9hrhRz+kDA==","signature_status":"signed_v1","signed_at":"2026-05-18T02:45:05.699300Z","signed_message":"canonical_sha256_bytes"},"source_id":"2601.22792","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:258683f0f5f566345d08a5112953414e3caadbc96dced08d31bcfb2f45aa0383","sha256:fdf9cd8cf4e030cd7ff3712880593c6c5c582f4e2659a8a81fe43a016fd4d66a"],"state_sha256":"d47223d6ad669098123f150d60c9bb20189865f7ab1592f168789982c45c6073"}