Tasks and inputs
Nemotron 3 Diarization is listed for Speaker Diarization. It accepts audio and produces speaker segments. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Oct 8, 2026
Nemotron 3 Diarization identifies who spoke when in audio, with streaming or offline operation and support for up to eight speakers.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Nemotron 3 Diarization identifies who spoke when in audio, with streaming or offline operation and support for up to eight speakers.
Nemotron 3 Diarization produces speaker-attributed time segments rather than a word transcript. It supports up to eight speakers and orders output channels by each speaker’s first arrival. Combine it with an ASR model when an application needs both words and a speaker timeline.
The same checkpoint supports several input-buffer settings. Its speaker cache preserves identity from earlier chunks while a recent-frame queue supplies local context. The publisher distinguishes input-buffer latency from computation time; a short buffer setting does not by itself establish end-to-end response latency.
Use recordings where a speaker returns after a long gap, two people overlap and background noise resembles speech. Compare segment boundaries and identity consistency with a manually reviewed reference. Record errors around very short turns separately, since those can materially affect a meeting summary.
The documented interface accepts audio files, arrays and manifests at the configured sample rate. NeMo and supported Transformers paths have their own settings. The roughly 99M-parameter checkpoint uses OpenMDW 1.1 licensing. A language-model token context and ordinary text-output price are not meaningful specifications for this diarization task, so those fields remain unset.
Compare the model on the work you need to complete.
Nemotron 3 Diarization is listed for Speaker Diarization. It accepts audio and produces speaker segments. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under OpenMDW 1.1. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
pyannote
Speaker diarization from audio: speaker-labeled time segments, exclusive diarization and a separate handoff to transcription.