{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:JPWIWD7VFA2UQXMFBDRRWDQNAA","short_pith_number":"pith:JPWIWD7V","schema_version":"1.0","canonical_sha256":"4bec8b0ff52835485d8508e31b0e0d001f30651a3b6bede22ce259be5b44e0e7","source":{"kind":"arxiv","id":"2112.05820","version":3},"attestation_state":"computed","paper":{"title":"Building a great multi-lingual teacher with sparsely-gated mixture of experts for speech recognition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","eess.AS"],"primary_cat":"cs.CL","authors_text":"Devang Patel, Eric Sun, Felipe Cruz Salinas, Kenichi Kumatani, Linquan Liu, Robert Gmyr, Wei Zuo, Yu Shi","submitted_at":"2021-12-10T20:37:03Z","abstract_excerpt":"The sparsely-gated Mixture of Experts (MoE) can magnify a network capacity with a little computational complexity. In this work, we investigate how multi-lingual Automatic Speech Recognition (ASR) networks can be scaled up with a simple routing algorithm in order to achieve better accuracy. More specifically, we apply the sparsely-gated MoE technique to two types of networks: Sequence-to-Sequence Transformer (S2S-T) and Transformer Transducer (T-T). We demonstrate through a set of ASR experiments on multiple language data that the MoE networks can reduce the relative word error rates by 16.3% "},"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":"2112.05820","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-12-10T20:37:03Z","cross_cats_sorted":["cs.AI","cs.LG","eess.AS"],"title_canon_sha256":"5d981dba03b6b1ecf5ade625e0d79020127a617592f99d061b53f459cc47dea8","abstract_canon_sha256":"23f9388de2208a0b5eb968678f5d74e4ec3ed1e634a5acfb67d450ef4d6021f1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:45:12.697609Z","signature_b64":"9ESm0oPj30Jb1IkUcB1qfsNyTAn7lFjSTnTv8mee/BeqWdkxx0PZxg4iBNxLmRzM8opB41h9hH4LfjABOooyDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4bec8b0ff52835485d8508e31b0e0d001f30651a3b6bede22ce259be5b44e0e7","last_reissued_at":"2026-07-05T03:45:12.697120Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:45:12.697120Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Building a great multi-lingual teacher with sparsely-gated mixture of experts for speech recognition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","eess.AS"],"primary_cat":"cs.CL","authors_text":"Devang Patel, Eric Sun, Felipe Cruz Salinas, Kenichi Kumatani, Linquan Liu, Robert Gmyr, Wei Zuo, Yu Shi","submitted_at":"2021-12-10T20:37:03Z","abstract_excerpt":"The sparsely-gated Mixture of Experts (MoE) can magnify a network capacity with a little computational complexity. In this work, we investigate how multi-lingual Automatic Speech Recognition (ASR) networks can be scaled up with a simple routing algorithm in order to achieve better accuracy. More specifically, we apply the sparsely-gated MoE technique to two types of networks: Sequence-to-Sequence Transformer (S2S-T) and Transformer Transducer (T-T). We demonstrate through a set of ASR experiments on multiple language data that the MoE networks can reduce the relative word error rates by 16.3% "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2112.05820","kind":"arxiv","version":3},"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/2112.05820/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":"2112.05820","created_at":"2026-07-05T03:45:12.697178+00:00"},{"alias_kind":"arxiv_version","alias_value":"2112.05820v3","created_at":"2026-07-05T03:45:12.697178+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2112.05820","created_at":"2026-07-05T03:45:12.697178+00:00"},{"alias_kind":"pith_short_12","alias_value":"JPWIWD7VFA2U","created_at":"2026-07-05T03:45:12.697178+00:00"},{"alias_kind":"pith_short_16","alias_value":"JPWIWD7VFA2UQXMF","created_at":"2026-07-05T03:45:12.697178+00:00"},{"alias_kind":"pith_short_8","alias_value":"JPWIWD7V","created_at":"2026-07-05T03:45:12.697178+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.15315","citing_title":"Tight Clusters Make Specialized Experts","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA","json":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA.json","graph_json":"https://pith.science/api/pith-number/JPWIWD7VFA2UQXMFBDRRWDQNAA/graph.json","events_json":"https://pith.science/api/pith-number/JPWIWD7VFA2UQXMFBDRRWDQNAA/events.json","paper":"https://pith.science/paper/JPWIWD7V"},"agent_actions":{"view_html":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA","download_json":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA.json","view_paper":"https://pith.science/paper/JPWIWD7V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2112.05820&json=true","fetch_graph":"https://pith.science/api/pith-number/JPWIWD7VFA2UQXMFBDRRWDQNAA/graph.json","fetch_events":"https://pith.science/api/pith-number/JPWIWD7VFA2UQXMFBDRRWDQNAA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA/action/storage_attestation","attest_author":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA/action/author_attestation","sign_citation":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA/action/citation_signature","submit_replication":"https://pith.science/pith/JPWIWD7VFA2UQXMFBDRRWDQNAA/action/replication_record"}},"created_at":"2026-07-05T03:45:12.697178+00:00","updated_at":"2026-07-05T03:45:12.697178+00:00"}