CatalogWantedMirrorsLedgerOperating StatementDocsStudio statusCard system
alt.freeweights.nemotron / Nemotron-3-Diarization-ONNX
From: onnx-community <huggingface.co/onnx-community>Newsgroups: alt.freeweights.nemotronSubject:

Nemotron-3-Diarization-ONNX

Date: 2026-11-15Commit: 353b6f8Signed: 2DC28CC7B26A283B ✓

Voice activity detection model. Onnx weights, 876 MB in all. Published by onnx-community under OpenMDW-1.1.

nemotron876MBNemotron-3-Diarization-ONNXonnxcustom terms
FactsParamsnot statedSize876 MBFormatsonnxTaskvoice activity detectionLicenseOpenMDW-1.1From sourcecustom termsread terms
# custom terms: read the license before you download
$ fw get onnx-community/Nemotron-3-Diarization-ONNX
# manifest sha256 88718778a399… · 876 MB · 15 files
$ fw verify ./Nemotron-3-Diarization-ONNX onnx-community/Nemotron-3-Diarization-ONNX
From sourceHosted (signed)onnx-community/Nemotron-3-Diarization-ONNXLast check: 2026-11-15 00:24 UTC
Run at homeGet help

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Quantized of Nemotron-3-Diarization.

Works with

Run2Convert3Scan1

Runs with

Ranked by fit to this model (onnx), then by GitHub stars.

UltralyticsOpenVINO

Convert with

ONNXOptimumOlive

Scan with

ModelScan
OverviewModel cardFilesUse itLicenseProvenanceModel tree

Overview

Size
876 MB
Weights
73.5 MB (smallest of 5 ONNX variants)
Parameters
not stated
Formats
onnx
Task
voice activity detection
License
OpenMDW-1.1 (custom)
Files
15, each with a sha256
Publisher
onnx-community
Added
2026-11-15
Section
alt.freeweights.nemotron
Lineage

Original release from the publisher.

Base model: nvidia/Nemotron-3-Diarization

Where these bytes come from
Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources
BadgeHostLast checkResult
From sourcehuggingface.co2026-11-15 00:24 UTCpassed

Model card

From the publisher at commit 353b6f8. Shown as plain text with links; images open at the publisher.

Nemotron 3 Diarization#

Description#

Nemotron 3 Diarization is an open-weight speaker diarization model designed to determine "who spoke when" in real-world audio. It supports both streaming and offline inference and handles up to eight speakers. Following Sortformer [\[1\]](#ref-1), the model resolves speaker permutation by ordering its output channels according to each speaker's first arrival in the input audio.

For streaming inference, the model adopts the Arrival-Order Speaker Cache (AOSC) and FIFO queue introduced in Streaming Sortformer [\[2\]](#ref-2). The AOSC retains speaker information from earlier chunks to preserve speaker identities over time, while the FIFO queue provides recent frame context for each processing step.

A single checkpoint supports input buffer latency as low as 80 ms for latency-critical applications, although the lowest recommended configuration is 0.32 s; the offline-style configuration uses a 30.4 s input buffer. The output frame resolution is configurable in multiples of 10 ms. With chunked inference, the maximum audio duration is not limited.

This model is ready for commercial or non-commercial use.

Nemotron 3 Diarization HuggingFace Blog

Release Date#

September 23, 2026

Live Action Demo Page#

Hugging Face Spaces Demo: Nemotron-Diarization with Streaming ASR Nemotron-Diarization with Streaming ASR Demo

What does max 8-speaker diarization mean ?#

How to use this model#

Coming soon...

Model Architecture#

Architecture Type: Transformer.

Network Architecture:

  • A 31-layer Transformer encoder with Rotary Positional Embeddings (RoPE).
  • 10 ms Mel-spectrogram input features downsampled by a factor of eight through feature stacking, producing an 80 ms encoder frame rate.
  • A Conv1D layer above the encoder that upsamples predictions to the 10 ms input-feature resolution.
  • AOSC and a FIFO queue for streaming inference.

Number of model parameters: 100M (1.0 × 10⁸).

Input#

Input Type: Audio.

Input Format: 16 kHz, single-channel audio in .wav, .flac, .opus, or .mp3 format.

Input Parameters: One-dimensional, 16 kHz, single-channel audio waveform.

Maximum Duration: Not limited when chunked inference is used.

Other Properties Related to Input: Input audio must be sampled at 16 kHz and is converted into 10 ms Mel-spectrogram features. Chunked inference supports recordings without a fixed maximum duration.

Output#

Output Type: Other: Numerical tensor.

Output Format: Float Tensor.

Output Parameters: A two-dimensional tensor with shape [T, 8], where T is the number of output frames and each value is a per-speaker activity probability in the range [0, 1].

Output Frame Resolution: The default frame stride is 10 ms and can be configured to any multiple of 10 ms, such as 30 ms, 80 ms, or 240 ms.

Speaker Ordering: The eight speaker channels are ordered by the speakers' arrival time in the input audio.

Derived Output: Speaker activity probabilities can be converted into start time, end time, and a generic speaker label, for example ["speaker1", 0.51, 12.62].

Other Properties Related to Output: The output frame resolution is configurable in multiples of 10 ms. Speaker channels are ordered by first arrival, and probabilities can be postprocessed into generic speaker labels with start and end timestamps.

Software Integration#

Runtime Engine(s): NeMo Framework v3.0.

Supported Hardware Microarchitecture Compatibility#

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems.

Ampere NVIDIA GPUs:

  • GeForce RTX 3090 Ti, RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070, RTX 3060 Ti, RTX 3060, and RTX 3050.
  • NVIDIA RTX A6000, RTX A5500, RTX A5000, RTX A4500, RTX A4000, RTX A2000, RTX A1000, and RTX A400.
  • NVIDIA A100 PCIe, A100 SXM, A30, A40, A16, A10, and A2.

Ampere NVIDIA Workstations: NVIDIA DGX Station A100.

Ada Lovelace NVIDIA GPUs:

  • GeForce RTX 4090, RTX 4080, RTX 4070 Ti, RTX 4070, RTX 4060 Ti, RTX 4060, and RTX 4050.
  • NVIDIA RTX 6000 Ada, RTX 5000 Ada, RTX 4500 Ada, RTX 4000 Ada, RTX 4000 SFF Ada, and RTX 2000 Ada.
  • NVIDIA L4, L40, and L40S.

Blackwell NVIDIA GPUs:

  • GeForce RTX 5090, RTX 5080, RTX 5070 Ti, RTX 5070, RTX 5060 Ti, RTX 5060, and RTX 5050.
  • RTX PRO 6000 Blackwell Workstation Edition, RTX PRO 6000 Blackwell Max-Q Workstation Edition, RTX PRO 5000 Blackwell, RTX PRO 4500 Blackwell Workstation Edition, RTX PRO 4000 Blackwell, RTX PRO 4000 Blackwell SFF Edition, and RTX PRO 2000 Blackwell.
  • RTX PRO 6000 Blackwell Server Edition and RTX PRO 4500 Blackwell Server Edition.
  • NVIDIA B200, B300, GB200, and GB300.

Blackwell NVIDIA Workstations: NVIDIA DGX Spark and NVIDIA DGX Station.

Hopper NVIDIA GPUs: NVIDIA H100 PCIe, H100 SXM, H100 NVL, H200 SXM, H200 NVL, and GH200.

Preferred/Supported Operating System(s):

  • Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version#

Nemotron 3 Diarization (General Access).

The model can be integrated into conversational AI systems through NeMo Framework inference pipelines to produce per-speaker activity probabilities or postprocessed generic speaker labels and timestamps.

Training and Evaluation Datasets#

Training Datasets#

The model was trained on a combination of real conversations and multi-talker audio mixtures with 1-8 speakers simulated using the FastMSS toolkit [\[3\]](#ref-3).

Real conversations: Total duration is around 10,000 hours.

DatasetLanguageSplit or description
Fisher EnglishEnglishTraining Part 1 and Part 2
AMI Meeting CorpusEnglishTrain and development; force-aligned [\[4\]](#ref-4)
ICSIEnglishFull
VoxConverse v0.3MultilingualDevelopment and test
AISHELL-4MandarinTrain
Third DIHARD ChallengeMultilingualDevelopment
2000 NIST Speaker Recognition EvaluationMultilingualCALLHOME Part 1
AliMeeting Mandarin CorpusMandarinTrain; force-aligned [\[4\]](#ref-4)
DiPCo — Dinner Party CorpusEnglishDevelopment
NOTSOFAR1EnglishTrain and development; force-aligned [\[3\]](#ref-3)
DISPLACE 2024English, Hindi, Kannada, Telugu, BengaliDevelopment and evaluation
DISPLACE-M 2026Hindi, KannadaDevelopment 1, 2, and 3
David AI — [D2] MultispeakerEnglishLicensed under agreement; 3-4 speakers 1,000 hours
YODAS-v2MultilingualPseudo-labeled 5,000-hour subset

Data used to simulate multi-talker audio mixtures: Total duration of single-speaker audio recordings is around 28,000 hours.

DatasetLanguageSplit or description
LibriSpeechEnglishTrain-960h
AMI Meeting Corpus (individual headsets)EnglishTrain and development
AliMeeting Mandarin Corpus (individual headsets)MandarinTrain
Fisher EnglishEnglishTraining Part 1 and Part 2
David AI — [D1] Chit ChatEnglishLicensed under agreement
David AI — [D2] MultispeakerEnglishLicensed under agreement
David AI — [D6a] PodcastEnglishLicensed under agreement
David AI — [D6b] AdviceEnglishLicensed under agreement
David AI — [D7] Expert AssistantEnglishLicensed under agreement
David AI — [D12] Human Transcripts21 languagesLicensed under agreement
MUSAN noisesNot applicableNoises for augmentation

Multi-talker audio mixtures used in training:

  • Librispeech: 6,694 hours
  • AMI: 6,743 hours
  • AliMeeting: 6,745 hours
  • Fisher English: 6,755 hours
  • David AI English: 36,458 hours
  • David AI Multilingual: 19,216 hours

Properties: Approximately 10,000 hours of real conversations plus 82,611 hours of simulated multi-talker audio mixtures. The data modality is audio and includes conversational speech, telephone calls, meetings, podcasts, noise augmentation, and synthetic mixtures. Languages include English, Mandarin, Hindi, Kannada, Telugu, Bengali, and other languages represented in the multilingual sources. Voice recordings may constitute personal data.

Evaluation Dataset#
DatasetLanguageSpeakersRecordingsDescriptionLabels
DIHARD III EvalMultilingual1–91–4 speakers: 219

5–9 speakers: 40 Total: 259 | 11-domain benchmark | Original | | CALLHOME-Part2 | Multilingual | 2–6 | 2: 148 3: 74 4: 20 5: 5 6: 3 Total: 250 | Telephonic speech | Original |

| AliMeeting Test Near | Mandarin | 2–4 | 20 | Meetings, mix of headset microphones | Forced alignment [\[4\]](#ref-4) |

| AliMeeting Test Far | Mandarin | 2–4 | 20 | Meetings, far-field conditions | Forced alignment [\[4\]](#ref-4) |

| AMI Test MHM | English | 3–4 | 16 | Meetings, mix of headset microphones | Forced alignment [\[4\]](#ref-4) |

| AMI Test SDM | English | 3–4 | 16 | Meetings, far-field single-channel conditions | Forced alignment [\[4\]](#ref-4) |

| NOTSOFAR1 Eval MHM | English | 3–7 | 3–4 speakers: 70 5–7 speakers: 90 Total: 160 | Meetings, mix of headset microphones | Forced alignment [\[3\]](#ref-3) | | NOTSOFAR1 Eval SC | English | 3–7 | 3–4 speakers: 70 5–7 speakers: 90 Total: 160 | Meetings, far-field single-channel conditions | Forced alignment [\[3\]](#ref-3) |

Properties: 901 condition-specific audio recordings comprising real-world multilingual conversational speech captured under telephone, meeting, near-field, far-field, and multi-microphone conditions. The data may contain personal data in the form of voice recordings. Languages include English, Mandarin, and other languages represented in the multilingual benchmarks.

Training#

The model training was initialized with a Transformer-based NEST [\[5\]](#ref-5) SSL checkpoint. Training was performed on 8 nodes of 8×NVIDIA A100-SXM4-80GB GPUs in two stages.

  1. Offline training on simulated data only using the example script and base config.
  2. Streaming fine-tuning on a combination of real conversations and simulated mixtures using the example script and base config.

Performance Evaluation#

Important.

Use the published reference labels to reproduce these results!#

The DER scores reported in this model card were computed using the exact reference annotations identified in the Labels column above. Reference RTTMs are part of the evaluation protocol: changing the reference labels changes the measured result.

We use forced-alignment based reference labels for AMI, AliMeeting and NOTSOFAR1 because the original segment-level annotations were created primarily for transcription rather than frame-accurate diarization evaluation. These annotations may label substantial within-segment silence as speech, thereby overestimating reference speaker activity. When used for DER scoring, they can inflate missed-speech error by penalizing a diarization system for correctly predicting non-speech during those intervals. Forced alignment provides more precise speech boundaries and therefore a more appropriate and interpretable reference for frame-level diarization evaluation. Please refer to [\[4\]](#ref-4) for a detailed discussion of this annotation issue and the forced-alignment methodology.

The Forced alignment links above point to the public repositories containing the reference RTTM files used for evaluation.

Results obtained using different reference labels constitute a different evaluation protocol and are not directly comparable with the numbers reported here. Before reporting a reproduction discrepancy, score the same model outputs using the linked reference RTTMs, the listed dataset split, and the collar and overlap settings specified in the Metrics section.

Baseline#

nvidia/diar_streaming_sortformer_4spk-v2.1

Baseline model's latency configurations:

ConfigurationLatencySPKCACHE_LENFIFO_LENCHUNK_LENRIGHT_CONTEXTUPDATE_PERIOD
Very high latency (offline)30.4 s1884034040300
Low latency1.04 s18818867144
Ultra-low latency0.32 s18818831144
Metrics#

Diarization Error Rate (DER): The primary metric for diarization performance, consisting of false alarm (FA), missed speech (Miss), and speaker confusion (Conf).

  • All evaluations include overlapping speech.
  • Collar tolerance is 0 s for DIHARD III Eval, AliMeeting Test, AMI Test and NOTSOFAR1 Eval.
  • Collar tolerance is 0.25 s for CALLHOME-part2.

Speaker Counting Accuracy (SCA): 1 when the predicted and ground-truth speaker counts are equal; otherwise 0. This metric does not capture the magnitude of a counting error.

Speaker Counting Mean Absolute Error (MAE): |predicted speaker count - ground-truth speaker count|. This metric is more informative than SCA because it reflects magnitude of speaker counting error.

Real-Time Factor Speedup (RTFx): total audio duration / total processing time.

Performance Evaluation Results#

Results reported below were obtained using the NeMo example script e2e_diarize_speech.py.

DIHARD III#

| Model | Latency | DER ↓ (1–4 spk) | DER ↓ (5–9 spk) | DER ↓ (full) | SCA ↑ (full) | MAE ↓ (full) |

| ------------------------------------- | ------- | ------------------ | ------------------ | --------------- | --------------- | --------------- |

| diar_streaming_sortformer_4spk-v2.1 | 30.4 s | 13.98 | 40.21 | 19.09 | 75.29 | 0.5135 |

| diar_streaming_sortformer_4spk-v2.1 | 1.04 s | 14.33 | 41.39 | 19.60 | 69.50 | 0.5483 |

| diar_streaming_sortformer_4spk-v2.1 | 0.32 s | 14.37 | 42.71 | 19.85 | 66.80 | 0.5869 |

| Nemotron-3-Diarization | 30.4 s | 9.13 | 27.58 | 12.73 | 81.47 | 0.2664 |

| Nemotron-3-Diarization | 1.04 s | 9.47 | 28.65 | 13.18 | 76.83 | 0.3243 |

| Nemotron-3-Diarization | 0.64 s | 9.44 | 29.16 | 13.28 | 77.22 | 0.3205 |

| Nemotron-3-Diarization | 0.32 s | 9.69 | 29.49 | 13.55 | 76.45 | 0.3282 |

CALLHOME-Part2#

| Model | Latency | DER ↓ (2 spk) | DER ↓ (3 spk) | DER ↓ (4 spk) | DER ↓ (5 spk) | DER ↓ (6 spk) | DER ↓ (full) | SCA ↑ (full) | MAE ↓ (full) |

| ------------------------------------- | ------- | ---------------- | ---------------- | ---------------- | ---------------- | ---------------- | --------------- | --------------- | --------------- |

| diar_streaming_sortformer_4spk-v2.1 | 30.4 s | 5.68 | 10.41 | 12.36 | 21.00 | 21.15 | 10.32 | 84.40 | 0.1720 |

| diar_streaming_sortformer_4spk-v2.1 | 1.04 s | 6.83 | 11.26 | 13.49 | 21.67 | 23.82 | 11.31 | 82.40 | 0.1880 |

| diar_streaming_sortformer_4spk-v2.1 | 0.32 s | 7.92 | 12.54 | 15.32 | 24.01 | 27.70 | 12.67 | 72.40 | 0.2920 |

| Nemotron-3-Diarization | 30.4 s | 5.98 | 9.26 | 11.03 | 15.82 | 15.80 | 9.10 | 91.60 | 0.0840 |

| Nemotron-3-Diarization | 1.04 s | 6.98 | 10.90 | 11.84 | 18.47 | 16.01 | 10.29 | 89.20 | 0.1080 |

| Nemotron-3-Diarization | 0.64 s | 7.21 | 11.17 | 12.10 | 19.31 | 16.70 | 10.66 | 88.40 | 0.1160 |

| Nemotron-3-Diarization | 0.32 s | 7.75 | 11.84 | 13.02 | 20.65 | 17.30 | 11.32 | 88.40 | 0.1160 |

AliMeeting Test Near#
ModelLatencyDER ↓SCA ↑MAE ↓
diar_streaming_sortformer_4spk-v2.130.4 s11.5780.000.20
diar_streaming_sortformer_4spk-v2.11.04 s12.4770.000.30
diar_streaming_sortformer_4spk-v2.10.32 s13.6870.000.30
Nemotron-3-Diarization30.4 s6.4090.000.10
Nemotron-3-Diarization1.04 s6.5985.000.15
Nemotron-3-Diarization0.64 s6.7485.000.15
Nemotron-3-Diarization0.32 s7.1980.000.20
AliMeeting Test Far#
ModelLatencyDER ↓SCA ↑MAE ↓
diar_streaming_sortformer_4spk-v2.130.4 s13.6995.000.05
diar_streaming_sortformer_4spk-v2.11.04 s15.5875.000.25
diar_streaming_sortformer_4spk-v2.10.32 s16.8565.000.35
Nemotron-3-Diarization30.4 s10.471000
Nemotron-3-Diarization1.04 s10.8095.000.05
Nemotron-3-Diarization0.64 s11.0385.000.15
Nemotron-3-Diarization0.32 s11.6085.000.15
AMI Test MHM#
ModelLatencyDER ↓SCA ↑MAE ↓
diar_streaming_sortformer_4spk-v2.130.4 s15.8193.750.0625
diar_streaming_sortformer_4spk-v2.11.04 s16.3693.750.0625
diar_streaming_sortformer_4spk-v2.10.32 s17.7793.750.0625
Nemotron-3-Diarization30.4 s9.2587.500.1250
Nemotron-3-Diarization1.04 s9.4881.250.1875
Nemotron-3-Diarization0.64 s9.6281.250.1875
Nemotron-3-Diarization0.32 s10.0581.250.1875
AMI Test SDM#
ModelLatencyDER ↓SCA ↑MAE ↓
diar_streaming_sortformer_4spk-v2.130.4 s21.4293.750.0625
diar_streaming_sortformer_4spk-v2.11.04 s21.7393.750.0625
diar_streaming_sortformer_4spk-v2.10.32 s23.8993.750.0625
Nemotron-3-Diarization30.4 s11.1487.500.1250
Nemotron-3-Diarization1.04 s12.8087.500.1250
Nemotron-3-Diarization0.64 s13.0687.500.1250
Nemotron-3-Diarization0.32 s12.9587.500.1250
NOTSOFAR1 MHM#

| Model | Latency | DER ↓ (3–4 spk) | DER ↓ (5–7 spk) | DER ↓ (full) | SCA ↑ (full) | MAE ↓ (full) |

| ------------------------------------- | ------- | ------------------ | ------------------ | --------------- | --------------- | --------------- |

| diar_streaming_sortformer_4spk-v2.1 | 30.4 s | 11.14 | 29.38 | 21.77 | 35.00 | 0.9375 |

| diar_streaming_sortformer_4spk-v2.1 | 1.04 s | 12.03 | 29.49 | 22.12 | 34.38 | 0.9437 |

| diar_streaming_sortformer_4spk-v2.1 | 0.32 s | 12.94 | 30.75 | 23.37 | 32.50 | 0.9625 |

| Nemotron-3-Diarization | 30.4 s | 5.25 | 7.86 | 6.77 | 93.75 | 0.0625 |

| Nemotron-3-Diarization | 1.04 s | 5.85 | 9.02 | 7.70 | 79.37 | 0.2062 |

| Nemotron-3-Diarization | 0.64 s | 6.07 | 9.39 | 7.99 | 79.37 | 0.2062 |

| Nemotron-3-Diarization | 0.32 s | 6.57 | 10.16 | 8.65 | 74.38 | 0.2687 |

NOTSOFAR1 SC#

| Model | Latency | DER ↓ (3–4 spk) | DER ↓ (5–7 spk) | DER ↓ (full) | SCA ↑ (full) | MAE ↓ (full) |

| ------------------------------------- | ------- | ------------------ | ------------------ | --------------- | --------------- | --------------- |

| diar_streaming_sortformer_4spk-v2.1 | 30.4 s | 19.67 | 38.42 | 30.49 | 33.12 | 0.9563 |

| diar_streaming_sortformer_4spk-v2.1 | 1.04 s | 20.58 | 39.84 | 31.81 | 33.12 | 0.9563 |

| diar_streaming_sortformer_4spk-v2.1 | 0.32 s | 22.14 | 40.80 | 32.95 | 31.87 | 0.9688 |

| Nemotron-3-Diarization | 30.4 s | 7.94 | 13.21 | 11.00 | 78.12 | 0.2188 |

| Nemotron-3-Diarization | 1.04 s | 8.96 | 15.44 | 12.77 | 60.62 | 0.4062 |

| Nemotron-3-Diarization | 0.64 s | 9.47 | 16.16 | 13.35 | 55.00 | 0.4625 |

| Nemotron-3-Diarization | 0.32 s | 10.28 | 17.61 | 14.53 | 55.00 | 0.4625 |

Inference Speed Evaluation#
ModelLatencyRTFx ↑ (batch_size=1) eager / compiledRTFx ↑ (batch_size=32) eager / compiled
diar_streaming_sortformer_4spk-v2.130.4 s874 / 14683204 / 2619
diar_streaming_sortformer_4spk-v2.11.04 s16 / 42193 / 136
diar_streaming_sortformer_4spk-v2.10.32 s8 / 21101 / 76
Nemotron-3-Diarization30.4 s1340 / 438512196 / 15113
Nemotron-3-Diarization1.04 s38 / 164581 / 865
Nemotron-3-Diarization0.64 s25 / 113391 / 579
Nemotron-3-Diarization0.32 s12.5 / 54199 / 292

Acceleration Engine: PyTorch backend through the NeMo Framework, with and without torch.compile().

Test Hardware: NVIDIA Blackwell RTX PRO 5000.

Precision: BF16.

License/Terms of Use#

Use of this model is governed by the OpenMDW License Agreement, version 1.1.

Deployment Geography#

Global

Use Case#

Nemotron 3 Diarization is intended for speaker diarization in live or recorded conversational audio, including meetings, calls, podcasts, and speech-recognition pipelines that need generic speaker labels and speaker timestamps.

References#

[1] Sortformer: A Novel Approach for Permutation-Resolved Speaker Supervision in Speech-to-Text Systems

[2] Streaming Sortformer: Speaker Cache-Based Online Speaker Diarization with Arrival-Time Ordering

[3] Mind the Gap: Impact of Synthetic Conversational Data on Multi-Talker ASR and Speaker Diarization

[4] Can We Really Repurpose Multi-Speaker ASR Corpus for Speaker Diarization?

[5] NEST: Self-supervised Fast Conformer as All-purpose Seasoning to Speech Processing Tasks

Ethical Considerations#

NVIDIA believes Trustworthy AI is a shared responsibility and has established policies and practices to enable development for a wide array of AI applications. Developers should evaluate the model with use-case-specific data and ensure that the complete system meets the requirements of the relevant industry and deployment context.

For more detailed information, see the Bias, Explainability, Safety & Security, and Privacy subcards.

Please report model quality, risk, security vulnerabilities, or NVIDIA AI concerns through the NVIDIA security reporting portal.

Files

15 files · 876 MBWeights 73.5 MB (smallest of 5 ONNX variants). Sizes and sha256 from the signed manifest.
Hide all 15 files with their sha256
FileSizesha256
.gitattributes1.83 KBba44193cf00eb237e2824cbfa212775d626f48f7c4caa3e630deda0908552db2
README.md31.3 KBf58e02164274cccb3003c766620e6c14885d1dc8243b4b91791703f72c3a3f62
config.json1.62 KBa758bf59867e27a7d92cf7a9149cae6dfc212b408b2ba74bfb87994708a803c5
onnx/model.onnx303 KB12a7c98fc1ee6ec98ea728c7ffd9fcf6b245f3c9c1ea9f7c2a7cc47c023e3f9d
onnx/model.onnx_data398 MBc293d9b5930eb9f6172f095ced052c0d1bbdbeb2594a115497583ca209b1dbd6
onnx/model_fp16.onnx352 KB286f56d64d80755eb0c84f9158c3620c9eccccb84ec53b48942c3795cf5a0c7b
onnx/model_fp16.onnx_data199 MBaffecf841c462d78c86b56a6eee6216b119809285bcbdb4d5c4d541810e9d317
onnx/model_q4.onnx360 KB3e5e93dfd8289770343c402079d36034faa4a2eac260f66115f85ba9a2de41b2
onnx/model_q4.onnx_data82.8 MB4513ed877d7cb83944bcb27c747221a7180e03f790d1c703e50af01792b7b8f4
onnx/model_q4f16.onnx408 KBe80ee1f1920f44c37d03bc65b1f1fd4f7450f5e1adbb0eaaccd3ab9442a543f7
onnx/model_q4f16.onnx_data73.1 MB3a9ba1de48ead635c04a117bae263f97fbd354358bca4197fb7184df7a5edf06
onnx/model_quantized.onnx364 KBfff7d18c7439c9fdc1c6c4dfec924cb42d3344264ca879780dfaf7ee886e6c1e
onnx/model_quantized.onnx_data120 MB002d7483e1c865c35c82220fdb378f185ff213c6d35922b38ae421c8ec72c338
preprocessor_config.json339 B71889e72aa30d334b83a05ba342c326a5ac6300920ca68eef6cabd01d4cc2292
processor_config.json623 B91726746788a024a2f0f2f699a6c3667f0d74e6a0e9288fc44fe3018e23af68e
VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.
Upstream commitDateContents
353b6f8ad2 pinned now2026-11-15 00:24 UTC15 files · 876 MB

Use it

# custom terms: read the license before you download
$ fw get onnx-community/Nemotron-3-Diarization-ONNX
# manifest sha256 88718778a399… · 876 MB · 15 files
$ fw verify ./Nemotron-3-Diarization-ONNX onnx-community/Nemotron-3-Diarization-ONNX
Terms that come with these files. OpenMDW 1.1: keep the agreement and the copyright notices with the files when you pass them on.
Custom license (custom). Read the license text first. Open this panel to show the commands once you accept its terms.
fw.pyChecks the signature and every sha256, tries each source in order, resumes.
$ python3 fw.py get https://freeweights.allternit.com/manifests/onnx-community/Nemotron-3-Diarization-ONNX.json --dir ./Nemotron-3-Diarization-ONNX $ python3 fw.py verify ./Nemotron-3-Diarization-ONNX ./onnx-community__Nemotron-3-Diarization-ONNX.json
hfThe Hugging Face CLI with one setting. Same repo id.
$ HF_ENDPOINT=https://freeweights.allternit.com hf download onnx-community/Nemotron-3-Diarization-ONNX --revision 353b6f8ad2cac3580e982d7fbdf0a010786b0406 --local-dir ./Nemotron-3-Diarization-ONNX
Run at home

License

Name
OpenMDW-1.1
Id
OpenMDW-1.1
Class
Custom terms
Redistribution
click-through
Listing
Hosted (signed)
Terms
OpenMDW 1.1: keep the agreement and the copyright notices with the files when you pass them on.
Decided
License as declared by the publisher. No person read it. 2026-11-15
Text in
README.md
Text sha256
f58e02164274cccb3003c766620e6c14885d1dc8243b4b91791703f72c3a3f62
Terms gate
Yes. The commands stay folded until you open them after reading the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From: onnx-community <huggingface.co/onnx-community>Newsgroups: alt.freeweights.nemotronSubject: Nemotron-3-Diarization-ONNXDate: 2026-11-15Commit: 353b6f8Signed: 2DC28CC7B26A283B ✓
Upstream
onnx-community/Nemotron-3-Diarization-ONNX
Pinned commit
353b6f8ad2cac3580e982d7fbdf0a010786b0406
Manifest
Nemotron-3-Diarization-ONNX.json · .minisig
Manifest sha256
88718778a399b1f4d38a6c03516dddc03db91972b31b4b475bbd775ba61effd8
Signature
minisign, Ed25519, key 2DC28CC7B26A283B
Created
2026-11-15 00:24 UTC
Generator
freeweights-manifest/0.1.0
Takedowns
None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/onnx-community/Nemotron-3-Diarization-ONNX.json $ curl -O https://freeweights.allternit.com/manifests/onnx-community/Nemotron-3-Diarization-ONNX.json.minisig $ minisign -Vm Nemotron-3-Diarization-ONNX.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq $ shasum -a 256 <file> # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

nvidia/Nemotron-3-Diarization · this model is a quantization of it

No model in the catalog names this one as its parent.

Get it
Read license, then download
$ HF_ENDPOINT=https://freeweights.allternit.com hf download onnx-community/Nemotron-3-Diarization-ONNX
Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest
minisign, key 2DC28CC7B26A283B. Nemotron-3-Diarization-ONNX.json. Provenance. Takedown history: none.
Source
huggingface.co/onnx-community/Nemotron-3-Diarization-ONNX

Related models

All of alt.freeweights.nemotron →
From: laion <huggingface.co/laion>Subject: a3-rl-laion_nemotron-gym-agent-calendar-80-8B
From: ServiceNow-AI <huggingface.co/ServiceNow-AI>Subject: Apriel-Nemotron-15b-Thinker
From: lmstudio-community <huggingface.co/lmstudio-community>Subject: Apriel-Nemotron-15b-Thinker-GGUF
From: nvidia <huggingface.co/nvidia>Subject: Nemotron-3-Diarization
From: nvidia <huggingface.co/nvidia>Subject: nemotron-3.5-asr-streaming-0.6b
From: bartowski <huggingface.co/bartowski>Subject: nvidia_OpenCodeReasoning-Nemotron-14B-GGUF