{"paper":{"title":"WaveNet: A Generative Model for Raw Audio","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"WaveNet generates raw audio waveforms by predicting each sample from all previous ones and yields more natural text-to-speech than prior systems.","cross_cats":["cs.LG"],"primary_cat":"cs.SD","authors_text":"Aaron van den Oord, Alex Graves, Andrew Senior, Heiga Zen, Karen Simonyan, Koray Kavukcuoglu, Nal Kalchbrenner, Oriol Vinyals, Sander Dieleman","submitted_at":"2016-09-12T17:29:40Z","abstract_excerpt":"This paper introduces WaveNet, a deep neural network for generating raw audio waveforms. The model is fully probabilistic and autoregressive, with the predictive distribution for each audio sample conditioned on all previous ones; nonetheless we show that it can be efficiently trained on data with tens of thousands of samples per second of audio. When applied to text-to-speech, it yields state-of-the-art performance, with human listeners rating it as significantly more natural sounding than the best parametric and concatenative systems for both English and Mandarin. A single WaveNet can captur"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"When applied to text-to-speech, it yields state-of-the-art performance, with human listeners rating it as significantly more natural sounding than the best parametric and concatenative systems for both English and Mandarin.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that human listener ratings of naturalness provide a reliable and unbiased measure of model quality, and that the autoregressive conditioning on prior samples plus speaker identity suffices to capture speaker characteristics without further mechanisms.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"WaveNet generates realistic raw audio using an autoregressive neural network with dilated convolutions, achieving state-of-the-art naturalness in speech synthesis for English and Mandarin.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"WaveNet generates raw audio waveforms by predicting each sample from all previous ones and yields more natural text-to-speech than prior systems.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ba7dd9419d135932cb469e7c957215415d0865d61dc54c69c33bc51dc3acdbb0"},"source":{"id":"1609.03499","kind":"arxiv","version":2},"verdict":{"id":"83a3d83a-2c58-40ac-b382-20b408abe767","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T20:23:22.651711Z","strongest_claim":"When applied to text-to-speech, it yields state-of-the-art performance, with human listeners rating it as significantly more natural sounding than the best parametric and concatenative systems for both English and Mandarin.","one_line_summary":"WaveNet generates realistic raw audio using an autoregressive neural network with dilated convolutions, achieving state-of-the-art naturalness in speech synthesis for English and Mandarin.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that human listener ratings of naturalness provide a reliable and unbiased measure of model quality, and that the autoregressive conditioning on prior samples plus speaker identity suffices to capture speaker characteristics without further mechanisms.","pith_extraction_headline":"WaveNet generates raw audio waveforms by predicting each sample from all previous ones and yields more natural text-to-speech than prior systems."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/1609.03499/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":60,"sample":[{"doi":"","year":2015,"title":"Vocaine the vocoder and applications is speech synthesis","work_id":"13476cf6-ab84-4f09-8aa9-f713544d7c30","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1994,"title":"Mixture density networks","work_id":"c96b52e8-1df3-465c-9ee1-c4f716c37066","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs","work_id":"d2a48269-aa6d-452f-895c-aafd2ec1f435","ref_index":3,"cited_arxiv_id":"1412.7062","is_internal_anchor":false},{"doi":"","year":1942,"title":"The Vowel: I ts Nature and Structure","work_id":"711ccdf7-1189-439b-b389-16643b375e61","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1939,"title":"Remaking speech","work_id":"320464ee-11c6-44a3-9c4e-2960841e1a5e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":60,"snapshot_sha256":"78f1bca38f6ca865264d250c59a67b5ed8dffd6cfc250c611e8e51e27d0f01e0","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1521357bb85baf8caaf186adc4e4b3e61fa30c32cfca2574349b0cef7c6fae4b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}