{"rewrite":{"id":"r_cd34ad30b678e33fb3a6ad85","clusterId":"c_05a068e84622de8277ca2e3e","slug":"nvidia-releases-nemotronlabs-re-use-noise-removal-model-as-open-weights","model":"deepseek-v4-1-flash","headline":"NVIDIA Releases NemotronLabs-RE-USE Noise Removal Model as Open Weights","summary":"NVIDIA has released NemotronLabs-RE-USE, an AI model that strips noise from audio files and outputs WAV. The model has 9.61 million parameters and is built with multilingual support in mind, removing noise regardless of language. In files where several speakers overlap, the main speaker is prioritized. It ships under the OpenMDW-1.1 license with weights on Hugging Face and a browser demo. GIGAZINE cites posted examples converting in-car recordings mixed with road noise, and audio mixed with speaker echo, into clear audio.","whyItMatters":"A 9.61-million-parameter noise remover released as open weights gives content creators and audio-AI training pipelines a free, self-hostable cleanup step, and the browser demo lets people test it before installing anything.","webCardHtml":"\u003cp\u003eThe release includes a demo app hosted on Hugging Face that runs in a browser, so the model can be tried without setting up a local environment. GIGAZINE points to posted examples where in-car recordings mixed with road noise, and audio carrying speaker echo, come out clear after processing. The parameter count is small by the standards of current generative audio, which matters for anyone running the model on modest hardware rather than through an API. A post on X from a user identified as Yusuke describes the results on Japanese audio; the same post notes quality drops when a large amount of noise is added.\u003c/p\u003e","blueskyPost":"NemotronLabs-RE-USE runs at 9.61 million parameters, small enough for NVIDIA to ship open weights plus a browser demo. The practical barrier to audio cleanup drops from subscription tools to a free download.","twitterPost":"NVIDIA's NemotronLabs-RE-USE delivers noise removal at 9.61 million parameters with open weights and a browser demo, so the barrier to cleanup is now a download.","threadsPost":"NemotronLabs-RE-USE is 9.61 million parameters, which is why NVIDIA can post the weights on Hugging Face and a browser demo alongside them. Audio cleanup stops being a paid subscription feature and becomes a file you run yourself.","newsletterBlurb":"NVIDIA has released NemotronLabs-RE-USE, an open model that removes noise from audio files and returns WAV. It has 9.61 million parameters, is designed for multilingual input, and prioritizes the main speaker when voices overlap. The weights are on Hugging Face under the OpenMDW-1.1 license, and a browser demo is live.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261102-nvidia-nemotronlabs-re-use/\",\"title\":\"NVIDIA Releases \\\"NemotronLabs-RE-USE,\\\" an AI That Removes Noise From Audio Files to Produce Clear Audio\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5616,"outputTokens":778,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1793606475,"createdAt":"2026-11-02T07:54:49.000Z","publishedAt":"2026-11-02T07:58:00.000Z","updatedAt":"2026-11-02T07:58:00.000Z"},"cluster":{"id":"c_05a068e84622de8277ca2e3e","canonicalTitle":"音声ファイルから雑音を除去してクリアな音声にできるAI「NemotronLabs-RE-USE」をNVIDIAが公開","representativeArticleId":"a_64d06f1f483e573e27807733","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"NemotronLabs-RE-USE\"],\"studios\":[\"NVIDIA\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-11-02T07:09:00.000Z","lastSeenAt":"2026-11-02T07:09:00.000Z","updatedAt":"2026-11-02T07:58:00.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261102-nvidia-nemotronlabs-re-use/","title":"音声ファイルから雑音を除去してクリアな音声にできるAI「NemotronLabs-RE-USE」をNVIDIAが公開"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["NemotronLabs-RE-USE"],"studios":["NVIDIA"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["NVIDIA released the NemotronLabs-RE-USE noise removal model on November 2, 2026.","NemotronLabs-RE-USE has 9.61 million parameters and outputs audio in WAV format.","The model is designed to remove noise regardless of language and prioritizes the main speaker in mixed audio.","NemotronLabs-RE-USE is distributed under the OpenMDW-1.1 license on Hugging Face.","A browser demo of NemotronLabs-RE-USE is available on Hugging Face Spaces."],"pageLinks":{"healthSlugs":["nemotronlabs-re-use","nvidia"],"storySlugs":[],"entities":[{"slug":"nemotronlabs-re-use","supported":0,"disputed":0,"contradicted":0,"unverified":9,"overall":"unverified","story_count":1,"has_dossier":false},{"slug":"nvidia","supported":6,"disputed":0,"contradicted":1,"unverified":31,"overall":"red","story_count":46,"has_dossier":true}],"stories":[],"entitiesOk":true,"storiesOk":true}}