{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:USZ4AUKFX5FEPFN2M5WLKVO75K","short_pith_number":"pith:USZ4AUKF","canonical_record":{"source":{"id":"2604.18105","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"eess.AS","submitted_at":"2026-04-20T11:21:06Z","cross_cats_sorted":["cs.CL","cs.SD"],"title_canon_sha256":"4e29e6fa0cdc1f5de6f0f4e3659be3b84fd492a0f0d5aa88938384fc564dfdbc","abstract_canon_sha256":"eb50ecfeb7b4d04e12a6667d85e729fa759cc3e133c6833632b1e1a61f76cb0c"},"schema_version":"1.0"},"canonical_sha256":"a4b3c05145bf4a4795ba676cb555dfea92a7d6f962d4df8198bd653ce9e134a2","source":{"kind":"arxiv","id":"2604.18105","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.18105","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"arxiv_version","alias_value":"2604.18105v2","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18105","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_12","alias_value":"USZ4AUKFX5FE","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_16","alias_value":"USZ4AUKFX5FEPFN2","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_8","alias_value":"USZ4AUKF","created_at":"2026-06-19T16:12:54Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:USZ4AUKFX5FEPFN2M5WLKVO75K","target":"record","payload":{"canonical_record":{"source":{"id":"2604.18105","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"eess.AS","submitted_at":"2026-04-20T11:21:06Z","cross_cats_sorted":["cs.CL","cs.SD"],"title_canon_sha256":"4e29e6fa0cdc1f5de6f0f4e3659be3b84fd492a0f0d5aa88938384fc564dfdbc","abstract_canon_sha256":"eb50ecfeb7b4d04e12a6667d85e729fa759cc3e133c6833632b1e1a61f76cb0c"},"schema_version":"1.0"},"canonical_sha256":"a4b3c05145bf4a4795ba676cb555dfea92a7d6f962d4df8198bd653ce9e134a2","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:54.407807Z","signature_b64":"aHCw6HNEztY7RqhAPGY6bOUe7UhOVdpA9Rc7Ci17FJkkor44bbzySFB7KpNk0ogcMWL8p8Uxz+F5hfFtdrzGBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a4b3c05145bf4a4795ba676cb555dfea92a7d6f962d4df8198bd653ce9e134a2","last_reissued_at":"2026-06-19T16:12:54.407420Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:54.407420Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.18105","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-19T16:12:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0wOvjwpfZW0yuIEwQc8C9Iej6gfQ+bP3VgbfHXmdWPcK7nTcATtaLhB/oF4j0QwcufWNDzOujtu8iNsT7yXmCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T10:31:40.648101Z"},"content_sha256":"24201a1a0e07665d73a71b709c8c96fdbef60594557d7749dfe74f8671d21fdc","schema_version":"1.0","event_id":"sha256:24201a1a0e07665d73a71b709c8c96fdbef60594557d7749dfe74f8671d21fdc"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:USZ4AUKFX5FEPFN2M5WLKVO75K","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"NIM4-ASR: Towards Efficient, Robust, and Customizable Real-Time LLM-Based ASR","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model.","cross_cats":["cs.CL","cs.SD"],"primary_cat":"eess.AS","authors_text":"Bowen Chen, Guang Qiu, Jiaqi Song, Jie Gao, Jie Wu, Junfeng Yuan, Kai Qiao, Ming Lei, Shengqing Liu, Xianliang Wang, Yi Zhang, Yuan Xie","submitted_at":"2026-04-20T11:21:06Z","abstract_excerpt":"Integrating large language models (LLMs) into automatic speech recognition (ASR) has become a mainstream paradigm in recent years. Although existing LLM-based ASR models demonstrate impressive performance on public benchmarks, their training remains predominantly data-driven, leaving key practical challenges insufficiently addressed -- particularly limited downward scalability in resource-constrained deployments and hallucinations under acoustically challenging conditions. To address these issues, we present NIM4-ASR, a production-oriented LLM-based ASR framework optimized for both efficiency "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"NIM4-ASR achieves state-of-the-art performance on multiple public benchmarks with merely 2.3B parameters, while substantially outperforming larger-scale competitors on internal benchmarks -- particularly in entity-intensive real-world scenarios.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the principled delineation of functional roles between encoder and LLM, combined with the specific multi-stage training (reformulated pre-training, iterative asynchronous SFT, ASR-specialized RL), will reliably align modules to their capability boundaries without introducing new failure modes.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"NIM4-ASR delivers SOTA ASR performance on public benchmarks using a 2.3B-parameter LLM with multi-stage training, real-time streaming, and million-scale hotword customization via RAG.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f3fbf301313a90a0e82033b26ba4c983b23193feac0d850fa07697ea0f567d46"},"source":{"id":"2604.18105","kind":"arxiv","version":2},"verdict":{"id":"8a9547a2-c813-4ae3-be25-1ad2389c3504","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T03:44:57.554538Z","strongest_claim":"NIM4-ASR achieves state-of-the-art performance on multiple public benchmarks with merely 2.3B parameters, while substantially outperforming larger-scale competitors on internal benchmarks -- particularly in entity-intensive real-world scenarios.","one_line_summary":"NIM4-ASR delivers SOTA ASR performance on public benchmarks using a 2.3B-parameter LLM with multi-stage training, real-time streaming, and million-scale hotword customization via RAG.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the principled delineation of functional roles between encoder and LLM, combined with the specific multi-stage training (reformulated pre-training, iterative asynchronous SFT, ASR-specialized RL), will reliably align modules to their capability boundaries without introducing new failure modes.","pith_extraction_headline":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.18105/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-20T04:26:08.443113Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"50324d32bd7a7c30e9185a3fdc1b8f58811043746b3b41aa6f833a518472cc6a"},"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":"8a9547a2-c813-4ae3-be25-1ad2389c3504"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-19T16:12:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4Brd+KnCCfvOxoq3Hl5u3MG+qZ61E6RtrRBxbzmKcOy5hcqmKM5DgE3ske9evDrD9ImcTMqzd/nqjkkykl/fBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T10:31:40.650358Z"},"content_sha256":"f05c40d03ce89544dd58930995f4fdf1fc89bb4ed0240f62a41caf12bd5920d2","schema_version":"1.0","event_id":"sha256:f05c40d03ce89544dd58930995f4fdf1fc89bb4ed0240f62a41caf12bd5920d2"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/bundle.json","state_url":"https://pith.science/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/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":"2026-08-06T10:31:40Z","links":{"resolver":"https://pith.science/pith/USZ4AUKFX5FEPFN2M5WLKVO75K","bundle":"https://pith.science/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/bundle.json","state":"https://pith.science/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/state.json","well_known_bundle":"https://pith.science/.well-known/pith/USZ4AUKFX5FEPFN2M5WLKVO75K/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:USZ4AUKFX5FEPFN2M5WLKVO75K","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":"eb50ecfeb7b4d04e12a6667d85e729fa759cc3e133c6833632b1e1a61f76cb0c","cross_cats_sorted":["cs.CL","cs.SD"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"eess.AS","submitted_at":"2026-04-20T11:21:06Z","title_canon_sha256":"4e29e6fa0cdc1f5de6f0f4e3659be3b84fd492a0f0d5aa88938384fc564dfdbc"},"schema_version":"1.0","source":{"id":"2604.18105","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.18105","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"arxiv_version","alias_value":"2604.18105v2","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18105","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_12","alias_value":"USZ4AUKFX5FE","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_16","alias_value":"USZ4AUKFX5FEPFN2","created_at":"2026-06-19T16:12:54Z"},{"alias_kind":"pith_short_8","alias_value":"USZ4AUKF","created_at":"2026-06-19T16:12:54Z"}],"graph_snapshots":[{"event_id":"sha256:f05c40d03ce89544dd58930995f4fdf1fc89bb4ed0240f62a41caf12bd5920d2","target":"graph","created_at":"2026-06-19T16:12:54Z","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":"NIM4-ASR achieves state-of-the-art performance on multiple public benchmarks with merely 2.3B parameters, while substantially outperforming larger-scale competitors on internal benchmarks -- particularly in entity-intensive real-world scenarios."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the principled delineation of functional roles between encoder and LLM, combined with the specific multi-stage training (reformulated pre-training, iterative asynchronous SFT, ASR-specialized RL), will reliably align modules to their capability boundaries without introducing new failure modes."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"NIM4-ASR delivers SOTA ASR performance on public benchmarks using a 2.3B-parameter LLM with multi-stage training, real-time streaming, and million-scale hotword customization via RAG."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model."}],"snapshot_sha256":"f3fbf301313a90a0e82033b26ba4c983b23193feac0d850fa07697ea0f567d46"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T04:26:08.443113Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2604.18105/integrity.json","findings":[],"snapshot_sha256":"50324d32bd7a7c30e9185a3fdc1b8f58811043746b3b41aa6f833a518472cc6a","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Integrating large language models (LLMs) into automatic speech recognition (ASR) has become a mainstream paradigm in recent years. Although existing LLM-based ASR models demonstrate impressive performance on public benchmarks, their training remains predominantly data-driven, leaving key practical challenges insufficiently addressed -- particularly limited downward scalability in resource-constrained deployments and hallucinations under acoustically challenging conditions. To address these issues, we present NIM4-ASR, a production-oriented LLM-based ASR framework optimized for both efficiency ","authors_text":"Bowen Chen, Guang Qiu, Jiaqi Song, Jie Gao, Jie Wu, Junfeng Yuan, Kai Qiao, Ming Lei, Shengqing Liu, Xianliang Wang, Yi Zhang, Yuan Xie","cross_cats":["cs.CL","cs.SD"],"headline":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"eess.AS","submitted_at":"2026-04-20T11:21:06Z","title":"NIM4-ASR: Towards Efficient, Robust, and Customizable Real-Time LLM-Based ASR"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.18105","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T03:44:57.554538Z","id":"8a9547a2-c813-4ae3-be25-1ad2389c3504","model_set":{"reader":"grok-4.3"},"one_line_summary":"NIM4-ASR delivers SOTA ASR performance on public benchmarks using a 2.3B-parameter LLM with multi-stage training, real-time streaming, and million-scale hotword customization via RAG.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A 2.3 billion parameter LLM-based ASR system reaches state-of-the-art accuracy by assigning distinct roles to its encoder and language model.","strongest_claim":"NIM4-ASR achieves state-of-the-art performance on multiple public benchmarks with merely 2.3B parameters, while substantially outperforming larger-scale competitors on internal benchmarks -- particularly in entity-intensive real-world scenarios.","weakest_assumption":"That the principled delineation of functional roles between encoder and LLM, combined with the specific multi-stage training (reformulated pre-training, iterative asynchronous SFT, ASR-specialized RL), will reliably align modules to their capability boundaries without introducing new failure modes."}},"verdict_id":"8a9547a2-c813-4ae3-be25-1ad2389c3504"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:24201a1a0e07665d73a71b709c8c96fdbef60594557d7749dfe74f8671d21fdc","target":"record","created_at":"2026-06-19T16:12:54Z","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":"eb50ecfeb7b4d04e12a6667d85e729fa759cc3e133c6833632b1e1a61f76cb0c","cross_cats_sorted":["cs.CL","cs.SD"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"eess.AS","submitted_at":"2026-04-20T11:21:06Z","title_canon_sha256":"4e29e6fa0cdc1f5de6f0f4e3659be3b84fd492a0f0d5aa88938384fc564dfdbc"},"schema_version":"1.0","source":{"id":"2604.18105","kind":"arxiv","version":2}},"canonical_sha256":"a4b3c05145bf4a4795ba676cb555dfea92a7d6f962d4df8198bd653ce9e134a2","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a4b3c05145bf4a4795ba676cb555dfea92a7d6f962d4df8198bd653ce9e134a2","first_computed_at":"2026-06-19T16:12:54.407420Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-19T16:12:54.407420Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"aHCw6HNEztY7RqhAPGY6bOUe7UhOVdpA9Rc7Ci17FJkkor44bbzySFB7KpNk0ogcMWL8p8Uxz+F5hfFtdrzGBQ==","signature_status":"signed_v1","signed_at":"2026-06-19T16:12:54.407807Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.18105","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:24201a1a0e07665d73a71b709c8c96fdbef60594557d7749dfe74f8671d21fdc","sha256:f05c40d03ce89544dd58930995f4fdf1fc89bb4ed0240f62a41caf12bd5920d2"],"state_sha256":"36617a5952d1a82b35bd23a30c0ccb4afd74dbcfc34cee395f596274bbc4ba69"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"T/w4q1wOZIb33XWRIYQl5pQR8aQmtO05Dk+LXc9BzuG/2XF20MbnoLb5BIwQI9GQqc90YHHcgGqX2tQ7b7q2Aw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-06T10:31:40.658600Z","bundle_sha256":"a5b685824ca5e9e0b1f43431f767b5ca28c6e39edb93816bff0b7eb47284e31f"}}