{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
    "title": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis",
    "summary": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.",
    "boundary": "Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.",
    "area": "On-device speech-generation systems",
    "topic": "Artificial intelligence",
    "year": 2026,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
    "target": {
      "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
      "version": 1,
      "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
    },
    "humans": [
      "Dongseong Hwang",
      "Prasanth Yadla",
      "Kaan Elgin",
      "Shifas Padinjaru Veettil",
      "Sivanand Achanta",
      "Dipjyoti Paul",
      "Ramya Rasipuram",
      "Tyler Johnson",
      "Emad Soroush",
      "Chung-Cheng Chiu",
      "Zhifeng Chen"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2607-23811v1",
        "title": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
        "url": "https://arxiv.org/abs/2607.23811v1"
      },
      {
        "id": "cur20261015-support-hwang-et-al-shared-depth-audio-detokenizer-2026",
        "title": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis — official institutional account",
        "url": "https://machinelearning.apple.com/research/audio-synthesis-diffusion-transformers"
      }
    ],
    "points": "55000",
    "published_grade": {
      "id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
      "entry_id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
      "band": "B",
      "modifier": "+",
      "target": {
        "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
        "version": 1,
        "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
      },
      "what_changed": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.",
      "scope": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization. Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
        "version": 1,
        "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Dongseong Hwang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Prasanth Yadla",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Kaan Elgin",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Shifas Padinjaru Veettil",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Sivanand Achanta",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Dipjyoti Paul",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ramya Rasipuram",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Tyler Johnson",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Emad Soroush",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Chung-Cheng Chiu",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Zhifeng Chen",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
            "url": "https://arxiv.org/abs/2607.23811v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis — official institutional account",
            "url": "https://machinelearning.apple.com/research/audio-synthesis-diffusion-transformers"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.",
        "domain": "On-device speech-generation systems"
      },
      "sources": [
        {
          "citation": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
          "url": "https://arxiv.org/abs/2607.23811v1"
        },
        {
          "citation": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis — official institutional account",
          "url": "https://machinelearning.apple.com/research/audio-synthesis-diffusion-transformers"
        }
      ],
      "statement": "# Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis\n\nSeparated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.\n\n## Scientific scope\n\nMoshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nDongseong Hwang; Prasanth Yadla; Kaan Elgin; Shifas Padinjaru Veettil; Sivanand Achanta; Dipjyoti Paul; Ramya Rasipuram; Tyler Johnson; Emad Soroush; Chung-Cheng Chiu; Zhifeng Chen\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers](https://arxiv.org/abs/2607.23811v1)\n- [Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis — official institutional account](https://machinelearning.apple.com/research/audio-synthesis-diffusion-transformers)\n",
      "summary": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.",
      "title": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis"
    },
    "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
      "version": 1,
      "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:d4e16432695fa4d73f4862ac14cfef5399e051c22a13348ef15cc0902a961c60"
  },
  "target": {
    "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
    "version": 1,
    "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
    "entry_id": "hwang-et-al-shared-depth-audio-detokenizer-2026",
    "revision": 1,
    "title": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis",
    "context": "On-device speech-generation systems",
    "track": "Human-designed research method or resource",
    "band": "B",
    "modifier": "+",
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.",
    "scope": "Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization. Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.",
    "references": [
      {
        "title": "Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers",
        "url": "https://arxiv.org/abs/2607.23811v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis — official institutional account",
        "url": "https://machinelearning.apple.com/research/audio-synthesis-diffusion-transformers",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:80bea850-f4cb-4d18-aba3-99fbf442eb92",
      "version": 1,
      "content_hash": "sha256:81e26021b1c217b51ffbfdd569192c7a76ace74f4a17e6346293de5814767d8a"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "55000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}