Skip to main content

NVIDIAAudio modelOpen weights

Nemotron-3-Diarization

PopularityTop group about equally popular, 8 models

56out of 100new calculation

based on two data groups, mostly search, range 46 to 66

Summary

Nemotron-3-Diarization assigns audio segments to up to eight speakers. NVIDIA’s card describes streaming and offline modes, 10 millisecond output resolution, and the OpenMDW-1.1 license.

Nemotron-3-Diarization uses local open weights rather than a hosted transcription service. It returns speaker segments, not a spoken audio response.

GraduallyBench Audio

Nemotron-3-Diarization

Technical test
  • Only a technical test is available.
  • The selection of test inputs is limited.
  • The functional test does not assess content quality.
Show test details

Technical measurement

Local · German, English

8 valid outputs from 8 attempted and 8 planned test slots.

No further scored metric is available for this test.

WER compares words and CER compares characters with the reference. The speech output was re-transcribed with GPT Transcribe. Its exact backend revision is unconfirmed. Read speech and digital silence do not cover spontaneous speech or real background noise. Error rates can exceed 100 %. Where used, the test references were not independently reviewed by humans. Test coverage and methods differ by model. Each metric stands on its own; there is no combined overall score across tests. Data compiled on 11/10/2026. How we test · Download all Decision and Audio results as JSON

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 105 published benchmark results

Page 1 of 9

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

DAI-ASR-I18N German DER, mono-channel (Show measurement, test conditions, and source)
Source value
14.6150394441
Metric
German diarization error rate (DER), mono-channel
Unit
%
Benchmark version
DAI-ASR-I18N v1.1, public per-language results embedded in page snapshot retrieved 2026-10-17
Category
speaker diarization
Direction
Lower is better
Source type
independent-evaluator
Evaluator
David AI Research, DAI-ASR-I18N
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact checkpoint in DAI Table A1: nvidia/Nemotron-3-Diarization, self-hosted with NeMo's released offline profile (speaker cache 264, FIFO 40, chunk length 340, right context 40, cache update period 300 frames), without a language hint or supplied speaker count. The captured page result row is model=nemotron3_ga_diarize, metric=der, channel=mono, language=de. DAI-ASR-I18N v1.1 evaluates DER against forced-aligned references, with no collar and overlap retained; diarization-only predictions receive no extra segment post-processing. The public set provides about two evaluation hours per language across 21 languages; the German row's clip count and confidence interval are not published.
Context
The page does not publish a German clip count or confidence interval for this row. Its static all-language DER chart and embedded aggregate result differ; this record retains only the explicit German result.
Source David AI Research, DAI-ASR-I18N
VoiceLab SpokesBiz diarization error rate (Show measurement, test conditions, and source)
Source value
22.5
Metric
Diarization error rate (DER), lower is better
Unit
%
Benchmark version
VoiceLab SpokesBiz evaluation, report published 2026-10-12, Table 1
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 1 reports the exact row nvidia/Nemotron-3-Diarization on the manually transcribed and diarized Polish SpokesBiz conversational corpus. A model segment counted as a match when it overlapped a reference-word timestamp by at least 50%; the reported DER is 22.5%.
Context
The report does not publish the evaluated recording or word count for this model, a confidence interval, a pinned model-weight revision, or runtime settings. The corpus description of over 650 hours and 925 recordings is not an evaluated sample count.
Source VoiceLab AI
VoiceLab SpokesBiz word diarization error rate (Show measurement, test conditions, and source)
Source value
4.8
Metric
Word-level diarization error rate (WDER), lower is better
Unit
%
Benchmark version
VoiceLab SpokesBiz evaluation, report published 2026-10-12, Table 1
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 1 reports the exact row nvidia/Nemotron-3-Diarization on the manually transcribed and diarized Polish SpokesBiz conversational corpus. A model segment counted as a match when it overlapped a reference-word timestamp by at least 50%; the reported WDER is 4.8%.
Context
The report does not publish the evaluated recording or word count for this model, a confidence interval, a pinned model-weight revision, or runtime settings. The corpus description of over 650 hours and 925 recordings is not an evaluated sample count.
Source VoiceLab AI
VoiceLab SpokesBiz exact speaker-count prediction rate (Show measurement, test conditions, and source)
Source value
80.6
Metric
Exact speaker-count prediction rate, higher is better
Unit
%
Benchmark version
VoiceLab SpokesBiz evaluation, report published 2026-10-12, Table 2
Category
audio
Direction
Higher is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 2 compares predicted and reference speaker counts for the exact Nemotron-3 model on SpokesBiz; the report gives an 80.6% exactly-right rate.
Context
The table gives rounded category percentages, which sum to 99.9%; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
VoiceLab SpokesBiz speaker-count overestimation rate (Show measurement, test conditions, and source)
Source value
19.1
Metric
Speaker-count overestimation rate, lower is better
Unit
%
Benchmark version
VoiceLab SpokesBiz evaluation, report published 2026-10-12, Table 2
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 2 compares predicted and reference speaker counts for the exact Nemotron-3 model on SpokesBiz; the report gives a 19.1% too-many-speakers rate.
Context
The table gives rounded category percentages, which sum to 99.9%; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
VoiceLab SpokesBiz speaker-count underestimation rate (Show measurement, test conditions, and source)
Source value
0.2
Metric
Speaker-count underestimation rate, lower is better
Unit
%
Benchmark version
VoiceLab SpokesBiz evaluation, report published 2026-10-12, Table 2
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 2 compares predicted and reference speaker counts for the exact Nemotron-3 model on SpokesBiz; the report gives a 0.2% too-few-speakers rate.
Context
The table gives rounded category percentages, which sum to 99.9%; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
VoiceLab DiaBiz diarization error rate (Show measurement, test conditions, and source)
Source value
23.3
Metric
Diarization error rate (DER), lower is better
Unit
%
Benchmark version
VoiceLab DiaBiz evaluation, report published 2026-10-12, Table 3
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 3 evaluates the exact row nvidia/Nemotron-3-Diarization on DiaBiz, a phone-based agent-customer dialogue corpus. The original stereo channels were mixed and the evaluation assumed two speakers in each recording; the reported DER is 23.3%.
Context
The report does not publish the evaluated recording or word count for this model, a confidence interval, a pinned model-weight revision, or runtime settings. The corpus descriptor of 410 hours is not an evaluated sample count.
Source VoiceLab AI
VoiceLab DiaBiz word diarization error rate (Show measurement, test conditions, and source)
Source value
6.3
Metric
Word-level diarization error rate (WDER), lower is better
Unit
%
Benchmark version
VoiceLab DiaBiz evaluation, report published 2026-10-12, Table 3
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 3 evaluates the exact row nvidia/Nemotron-3-Diarization on DiaBiz, a phone-based agent-customer dialogue corpus. The original stereo channels were mixed and the evaluation assumed two speakers in each recording; the reported WDER is 6.3%.
Context
The report does not publish the evaluated recording or word count for this model, a confidence interval, a pinned model-weight revision, or runtime settings. The corpus descriptor of 410 hours is not an evaluated sample count.
Source VoiceLab AI
VoiceLab DiaBiz exact speaker-count prediction rate (Show measurement, test conditions, and source)
Source value
90.8
Metric
Exact speaker-count prediction rate, higher is better
Unit
%
Benchmark version
VoiceLab DiaBiz evaluation, report published 2026-10-12, Table 4
Category
audio
Direction
Higher is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 4 compares predicted and reference speaker counts for the exact Nemotron-3 model on DiaBiz; the evaluation mixed the stereo channels and assumed two speakers per recording. It reports a 90.8% exactly-right rate.
Context
The table gives rounded category percentages; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
VoiceLab DiaBiz speaker-count overestimation rate (Show measurement, test conditions, and source)
Source value
9.2
Metric
Speaker-count overestimation rate, lower is better
Unit
%
Benchmark version
VoiceLab DiaBiz evaluation, report published 2026-10-12, Table 4
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 4 compares predicted and reference speaker counts for the exact Nemotron-3 model on DiaBiz; the evaluation mixed the stereo channels and assumed two speakers per recording. It reports a 9.2% too-many-speakers rate.
Context
The table gives rounded category percentages; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
VoiceLab DiaBiz speaker-count underestimation rate (Show measurement, test conditions, and source)
Source value
0
Metric
Speaker-count underestimation rate, lower is better
Unit
%
Benchmark version
VoiceLab DiaBiz evaluation, report published 2026-10-12, Table 4
Category
audio
Direction
Lower is better
Source type
independent-evaluator
Evaluator
VoiceLab AI
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. VoiceLab Table 4 compares predicted and reference speaker counts for the exact Nemotron-3 model on DiaBiz; the evaluation mixed the stereo channels and assumed two speakers per recording. It reports a 0.0% too-few-speakers rate.
Context
The table gives rounded category percentages; no evaluated N or confidence interval is published, and no checkpoint or runtime is pinned.
Source VoiceLab AI
Soniqo Nemotron 3 INT8 MLX DER, Arabic (Show measurement, test conditions, and source)
Source value
16.54
Metric
Diarization error rate (DER), reference-speaker-time weighted with 0.25 s collar, Arabic
Unit
%
Benchmark version
Soniqo diarization page captured 2026-10-17T22:03:26Z; test version and run date not stated
Category
audio
Direction
Lower is better
Sample
12
Source type
vendor-reported
Evaluator
Soniqo Nemotron 3 speaker diarization benchmark
Status
active
Retrieved at
2026-10-18
Methodology
Vendor-reported result. Soniqo Nemotron 3 INT8 export, MLX backend, exact table language Arabic, 12 evaluated files in that language. The full internal set is 132 recordings across eight languages and 12.82 hours, with no failed files; sampleCount here is the table's per-language Files cell, not a speaker-turn or duration count. No reference speaker count is passed; automatic speaker counting and the package default postprocessing are used. DER is pooled by reference speaker time inside each recording's evaluation map, including overlap, with a 0.25 s collar. The source places this language in the real-conversation tier (broadcast, telephone and meeting audio).
Context
Soniqo publishes and operates this benchmark for its own shipped Apple Silicon runtimes, so the results are vendor-reported rather than a neutral independent reproduction. The page identifies the upstream as the final NVIDIA Nemotron 3 Diarization checkpoint but supplies no immutable checkpoint revision or converted-export hash, no Apple chip/OS or runtime version, and no benchmark run date. These observations apply only to the named Soniqo INT8 MLX or CoreML export with the package default postprocessing, not to unquantized NVIDIA model performance. No CI is stated. Soniqo explicitly separates real-conversation recordings from synthetic rails and says their results must not be compared across tiers.
Source Soniqo Nemotron 3 speaker diarization benchmark

Page 1 of 9

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

0 to 100, higher is more popular

20 of 54 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

Nemotron-3-DiarizationGemini 3.5 Transcribe 52Fish Audio S2 Pro 52Nova-3 51Phonon-2 51

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

All specifications

Architecture

Model class
Speaker diarization
Source: Model class
Checkpoint
nvidia/Nemotron-3-Diarization
Source: Checkpoint

Input and output

Input
Audio
Source: Input
Output
Speaker segments with time resolution
Source: Output

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Access
Local weights
Source: Access
License
OpenMDW-1.1
Source: License
Shutdown
None announced
no source

Further details

Speakers
Up to eight speakers in streaming and offline modes
Source: Speakers
Time resolution
Output resolution in 10 millisecond steps
Source: Time resolution

Sources and data date

Show all 19 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 12 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not published

immutable-model-weight-revision

Note

The results card names Nemotron-3-Diarization but does not bind these reported runs to an immutable model-weight revision.

Not published

confidence-intervals

Note

The card reports point DER values without confidence intervals.

Additional sourceNVIDIA Nemotron-3-Diarization model card (retrieved October 3, 2026) · NVIDIA Nemotron-3-Diarization model card · Editorial description reviewed October 3, 2026
Additional sourceNVIDIA Nemotron announcement (retrieved October 4, 2026)
Additional sourceNVIDIA Nemotron-3-Diarization model card (retrieved October 16, 2026)
Additional sourceDavid AI Research, DAI-ASR-I18N (retrieved October 17, 2026; October 19, 2026)
Additional sourceVoiceLab AI (retrieved October 17, 2026)
Additional sourceSoniqo Nemotron 3 speaker diarization benchmark (retrieved October 18, 2026)
Additional sourceArgmax OpenBench (retrieved October 18, 2026)
Additional sourceOmi-Health clinical diarization benchmark (retrieved October 19, 2026; October 24, 2026)
Additional sourceModel metadata (retrieved October 16, 2026)
Additional sourceModel metadata (retrieved October 18, 2026)
Additional sourceModel metadata (retrieved October 19, 2026)123
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0