--- title: NVIDIA Nemotron 3 Diarization: open-weight speaker... description: NVIDIA released Nemotron 3 Diarization, an open-weight model that identifies who spoke when in audio, working alongside ASR systems like Parakeet TDT, Canary,... canonical: https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: NVIDIA Nemotron 3 Diarization: open-weight speaker diarization for up to eight speakers | daily.dev og:description: NVIDIA released Nemotron 3 Diarization, an open-weight model that identifies who spoke when in audio, working alongside ASR systems like Parakeet TDT, Canary,... og:url: https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr og:image: https://api.daily.dev/og/posts/DEUFemRgr.png og:image:alt: NVIDIA Nemotron 3 Diarization: open-weight speaker diarization for up to eight speakers og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # NVIDIA Nemotron 3 Diarization: open-weight speaker diarization for up to eight speakers **[Collections](https://daily.dev/sources/collections)** · 2 min read · 1 upvotes · 0 comments ## Summary NVIDIA released Nemotron 3 Diarization, an open-weight model that identifies who spoke when in audio, working alongside ASR systems like Parakeet TDT, Canary, and Nemotron ASR. At around 100 million parameters, it runs on GPUs with as little as 4GB memory, handles up to eight speakers, and supports both streaming and offline modes. It ranks first on Voice Arena's Diarization-Bench with a 14.72% Diarization Error Rate, and replaces NVIDIA's four-speaker Streaming Sortformer baseline with a 41% relative DER reduction. It ships with a commercially permissive license and setup support via Argmax Pro SDK 3. ## Content NVIDIA released Nemotron 3 Diarization, a 100M-parameter open-weight model that figures out who said what and when in an audio recording. It supports up to eight speakers, handles overlapping speech, and works in both streaming and offline modes. The license is commercially permissive. The model currently sits at #1 on VoiceArena's Diarization-Bench across 12 systems, with a 14.72% Diarization Error Rate. One number worth understanding: that score assumes a 0ms collar (zero timing tolerance around speaker changes). Score the same output with a 250ms collar and DER drops to 4.29%. The collar effect is real, and it's worth knowing which scoring rules a benchmark uses before reading too much into the numbers. Compared to NVIDIA's previous Streaming Sortformer baseline (which was limited to four speakers), Nemotron 3 achieves roughly a 41% relative DER reduction at 1.04-second latency, and throughput jumps from 2,619x RTFx to 15,113x at 30.4-second latency. Latency is configurable, with operating points ranging from 30.4 seconds down to 0.32 seconds depending on your use case. For streaming, the model uses arrival-order speaker labeling with an Arrival-Order Speaker Cache and FIFO queue to maintain context. It integrates with ASR models like Parakeet TDT to produce speaker-attributed transcripts rather than just a raw transcript with no attribution. On the practical side: it runs on a GPU with 4GB of memory, which is a reasonable bar for most deployment scenarios. It's language-agnostic, so it doesn't need to understand what's being said to track who's saying it. It works alongside any ASR model, including Parakeet, Canary, or Nemotron ASR. The older Sortformer model it replaces had over 300,000 downloads in a single month, so there's clearly demand here. Argmax Pro SDK 3 has already added third-party integration. Demo outputs include speaker-attributed transcripts, per-speaker talk time and turn counts, and export to text or SRT format, all processed locally. ## Questions this post answers ### What is NVIDIA Nemotron 3 Diarization and how many speakers can it handle? Nemotron 3 Diarization is NVIDIA's open-weight speaker diarization model that identifies who spoke when in audio recordings, handling up to eight speakers compared to three or four in older models. At around 100 million parameters, it runs on GPUs with as little as 4GB of memory, supports overlapping speech, is language-agnostic, and works in streaming or offline modes under a commercially permissive license. _Teams building voice pipelines can follow diarization model releases like this one on daily.dev._ ### How does Nemotron 3 Diarization compare to NVIDIA's Sortformer model? Nemotron 3 Diarization replaces NVIDIA's previous four-speaker Streaming Sortformer baseline, which had over 300,000 downloads in a single month, achieving an average 41% relative Diarization Error Rate reduction at 1.04-second latency. Throughput reaches 15,113x real-time factor versus Sortformer's 2,619x at 30.4-second latency, and it supports up to eight speakers instead of three or four. _Anyone weighing an upgrade from Sortformer can track model comparisons like this on daily.dev._ ### What is the Diarization Error Rate of NVIDIA Nemotron 3 on Voice Arena's benchmark? Nemotron 3 Diarization ranks first on Voice Arena's Diarization-Bench with a 14.72% Diarization Error Rate, roughly a 24% relative improvement over the next-ranked system. It also exposes configurable latency operating points from 0.32 to 30.4 seconds, using an Arrival-Order Speaker Cache and FIFO queue for streaming context. _Developers benchmarking speech models can keep tabs on results like these via daily.dev._ --- Tags: [#open-source](https://daily.dev/tags/open-source), [#nvidia](https://daily.dev/tags/nvidia), [#speech-recognition](https://daily.dev/tags/speech-recognition) [View this post on daily.dev](https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"NVIDIA Nemotron 3 Diarization: open-weight speaker diarization for up to eight speakers","url":"https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr"},"datePublished":"2026-09-23T15:53:59.435Z","dateModified":"2026-09-23T23:46:55.773Z","description":"NVIDIA released Nemotron 3 Diarization, an open-weight model that identifies who spoke when in audio, working alongside ASR systems like Parakeet TDT, Canary,...","image":"https://i.ytimg.com/vi/PZuuOXNB3Vw/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/PZuuOXNB3Vw/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,nvidia,speech-recognition","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"NVIDIA Nemotron 3 Diarization: open-weight speaker diarization for up to eight speakers"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-nemotron-3-diarization-open-weight-speaker-diarization-for-up-to-eight-speakers-deufemrgr#faq","mainEntity":[{"@type":"Question","name":"What is NVIDIA Nemotron 3 Diarization and how many speakers can it handle?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3 Diarization is NVIDIA's open-weight speaker diarization model that identifies who spoke when in audio recordings, handling up to eight speakers compared to three or four in older models. At around 100 million parameters, it runs on GPUs with as little as 4GB of memory, supports overlapping speech, is language-agnostic, and works in streaming or offline modes under a commercially permissive license. Teams building voice pipelines can follow diarization model releases like this one on daily.dev."}},{"@type":"Question","name":"How does Nemotron 3 Diarization compare to NVIDIA's Sortformer model?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3 Diarization replaces NVIDIA's previous four-speaker Streaming Sortformer baseline, which had over 300,000 downloads in a single month, achieving an average 41% relative Diarization Error Rate reduction at 1.04-second latency. Throughput reaches 15,113x real-time factor versus Sortformer's 2,619x at 30.4-second latency, and it supports up to eight speakers instead of three or four. Anyone weighing an upgrade from Sortformer can track model comparisons like this on daily.dev."}},{"@type":"Question","name":"What is the Diarization Error Rate of NVIDIA Nemotron 3 on Voice Arena's benchmark?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3 Diarization ranks first on Voice Arena's Diarization-Bench with a 14.72% Diarization Error Rate, roughly a 24% relative improvement over the next-ranked system. It also exposes configurable latency operating points from 0.32 to 30.4 seconds, using an Arrival-Order Speaker Cache and FIFO queue for streaming context. Developers benchmarking speech models can keep tabs on results like these via daily.dev."}}]} ```