{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TNGIYICMVWZGTNQKA427AVMEZC","short_pith_number":"pith:TNGIYICM","schema_version":"1.0","canonical_sha256":"9b4c8c204cadb269b60a0735f05584c8b6de253c14b85a9c0e2a90b9f5b13454","source":{"kind":"arxiv","id":"2507.05911","version":1},"attestation_state":"computed","paper":{"title":"Differentiable Reward Optimization for LLM based TTS system","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","eess.AS"],"primary_cat":"cs.SD","authors_text":"Changfeng Gao, Shiliang Zhang, Zhihao Du","submitted_at":"2025-07-08T11:57:16Z","abstract_excerpt":"This paper proposes a novel Differentiable Reward Optimization (DiffRO) method aimed at enhancing the performance of neural codec language models based text-to-speech (TTS) systems. In contrast to conventional reinforcement learning from human feedback (RLHF) approaches applied to TTS, DiffRO directly compute the rewards based on neural codec tokens, rather than relying on synthesized audio. Furthermore, we employ the Gumbel-Softmax technique to render the reward function differentiable, thereby streamlining the RLHF training process. Additionally, we introduce a multi-task reward (MTR) model "},"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":"2507.05911","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.SD","submitted_at":"2025-07-08T11:57:16Z","cross_cats_sorted":["cs.AI","eess.AS"],"title_canon_sha256":"6d804c488979d11f55dc073fc4d9ea9e37cfbf8ffa4733a90607bee48debd178","abstract_canon_sha256":"2ededa7b24c991c49c6cb1008c684e3b7ef3b54fdce13e0639fca509678cb350"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:33:38.264667Z","signature_b64":"E/dtw9bIWQm8t3nliIOGgXnFtQ2WDRuO2PmmNhvgLAhSlA+R3AzLi1GVvru69qF1r2LMdSn8Qmb21ioJBG34Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9b4c8c204cadb269b60a0735f05584c8b6de253c14b85a9c0e2a90b9f5b13454","last_reissued_at":"2026-07-05T11:33:38.264092Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:33:38.264092Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Differentiable Reward Optimization for LLM based TTS system","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","eess.AS"],"primary_cat":"cs.SD","authors_text":"Changfeng Gao, Shiliang Zhang, Zhihao Du","submitted_at":"2025-07-08T11:57:16Z","abstract_excerpt":"This paper proposes a novel Differentiable Reward Optimization (DiffRO) method aimed at enhancing the performance of neural codec language models based text-to-speech (TTS) systems. In contrast to conventional reinforcement learning from human feedback (RLHF) approaches applied to TTS, DiffRO directly compute the rewards based on neural codec tokens, rather than relying on synthesized audio. Furthermore, we employ the Gumbel-Softmax technique to render the reward function differentiable, thereby streamlining the RLHF training process. Additionally, we introduce a multi-task reward (MTR) model "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.05911","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/2507.05911/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":"2507.05911","created_at":"2026-07-05T11:33:38.264155+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.05911v1","created_at":"2026-07-05T11:33:38.264155+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.05911","created_at":"2026-07-05T11:33:38.264155+00:00"},{"alias_kind":"pith_short_12","alias_value":"TNGIYICMVWZG","created_at":"2026-07-05T11:33:38.264155+00:00"},{"alias_kind":"pith_short_16","alias_value":"TNGIYICMVWZGTNQK","created_at":"2026-07-05T11:33:38.264155+00:00"},{"alias_kind":"pith_short_8","alias_value":"TNGIYICM","created_at":"2026-07-05T11:33:38.264155+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09234","citing_title":"End-to-End Training for Discrete Token LLM based TTS System","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC","json":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC.json","graph_json":"https://pith.science/api/pith-number/TNGIYICMVWZGTNQKA427AVMEZC/graph.json","events_json":"https://pith.science/api/pith-number/TNGIYICMVWZGTNQKA427AVMEZC/events.json","paper":"https://pith.science/paper/TNGIYICM"},"agent_actions":{"view_html":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC","download_json":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC.json","view_paper":"https://pith.science/paper/TNGIYICM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.05911&json=true","fetch_graph":"https://pith.science/api/pith-number/TNGIYICMVWZGTNQKA427AVMEZC/graph.json","fetch_events":"https://pith.science/api/pith-number/TNGIYICMVWZGTNQKA427AVMEZC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC/action/storage_attestation","attest_author":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC/action/author_attestation","sign_citation":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC/action/citation_signature","submit_replication":"https://pith.science/pith/TNGIYICMVWZGTNQKA427AVMEZC/action/replication_record"}},"created_at":"2026-07-05T11:33:38.264155+00:00","updated_at":"2026-07-05T11:33:38.264155+00:00"}