{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3VDZDIHWCRHMW542YAPNJWDDU4","short_pith_number":"pith:3VDZDIHW","schema_version":"1.0","canonical_sha256":"dd4791a0f6144ecb779ac01ed4d863a73948847bdf4f43df55f34b121d6ae818","source":{"kind":"arxiv","id":"2606.10029","version":1},"attestation_state":"computed","paper":{"title":"Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Georgii Aparin, Nikita Balagansky, Nikita Koriagin","submitted_at":"2026-06-08T18:09:37Z","abstract_excerpt":"Language models increasingly serve as the backbone of text-to-speech (TTS) systems, yet we understand little about the representations they build when text and generated speech tokens share a single residual stream. We train BatchTopK sparse autoencoders on the LM backbone of CosyVoice3 and introduce a modality-aware auto-interp pipeline that labels each feature from where it fires-text-prefix context, 1-second speech clips, or both. The recovered features are interpretable, spanning phonemes, laughter, accent prompts and speaker gender. Steering through the SAE latent space shows these featur"},"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":"2606.10029","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-08T18:09:37Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"e8eed618ae8badce98b6e89abdcdbab93878ac5b3a1a47b611501b2b6ede75b1","abstract_canon_sha256":"216e07ebcc5c05947f1be26909cd763a15d1d406103ce218a3314acd92e10cad"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T00:08:45.385069Z","signature_b64":"j+SfXrz3KeWq4dxv2B9KzUdbpbfe2TAwh99w3xdRoPgcUqYh8z0jQlldhy7en6GPDU9yLjjOY4lsCmEv29DuBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dd4791a0f6144ecb779ac01ed4d863a73948847bdf4f43df55f34b121d6ae818","last_reissued_at":"2026-06-10T00:08:45.384192Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T00:08:45.384192Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Georgii Aparin, Nikita Balagansky, Nikita Koriagin","submitted_at":"2026-06-08T18:09:37Z","abstract_excerpt":"Language models increasingly serve as the backbone of text-to-speech (TTS) systems, yet we understand little about the representations they build when text and generated speech tokens share a single residual stream. We train BatchTopK sparse autoencoders on the LM backbone of CosyVoice3 and introduce a modality-aware auto-interp pipeline that labels each feature from where it fires-text-prefix context, 1-second speech clips, or both. The recovered features are interpretable, spanning phonemes, laughter, accent prompts and speaker gender. Steering through the SAE latent space shows these featur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.10029","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/2606.10029/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":"2606.10029","created_at":"2026-06-10T00:08:45.384342+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.10029v1","created_at":"2026-06-10T00:08:45.384342+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.10029","created_at":"2026-06-10T00:08:45.384342+00:00"},{"alias_kind":"pith_short_12","alias_value":"3VDZDIHWCRHM","created_at":"2026-06-10T00:08:45.384342+00:00"},{"alias_kind":"pith_short_16","alias_value":"3VDZDIHWCRHMW542","created_at":"2026-06-10T00:08:45.384342+00:00"},{"alias_kind":"pith_short_8","alias_value":"3VDZDIHW","created_at":"2026-06-10T00:08:45.384342+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4","json":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4.json","graph_json":"https://pith.science/api/pith-number/3VDZDIHWCRHMW542YAPNJWDDU4/graph.json","events_json":"https://pith.science/api/pith-number/3VDZDIHWCRHMW542YAPNJWDDU4/events.json","paper":"https://pith.science/paper/3VDZDIHW"},"agent_actions":{"view_html":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4","download_json":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4.json","view_paper":"https://pith.science/paper/3VDZDIHW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.10029&json=true","fetch_graph":"https://pith.science/api/pith-number/3VDZDIHWCRHMW542YAPNJWDDU4/graph.json","fetch_events":"https://pith.science/api/pith-number/3VDZDIHWCRHMW542YAPNJWDDU4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4/action/storage_attestation","attest_author":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4/action/author_attestation","sign_citation":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4/action/citation_signature","submit_replication":"https://pith.science/pith/3VDZDIHWCRHMW542YAPNJWDDU4/action/replication_record"}},"created_at":"2026-06-10T00:08:45.384342+00:00","updated_at":"2026-06-10T00:08:45.384342+00:00"}