nemotron-3.5-asr-streaming-0.6b

Name: nemotron-3.5-asr-streaming-0.6b
Author: nvidia

by nvidia

600M multilingual streaming ASR with cache-aware FastConformer-RNNT

4.2Kdl/month

600Mparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

nvidia/nemotron-3.5-asr-streaming-0.6b

Feature URI

mixpeek://transcription@v1/nvidia_nemotron_35_asr_streaming_v1

Overview

Nemotron 3.5 ASR Streaming 0.6B is NVIDIA's multilingual streaming speech recognition model. The model card describes a 600M parameter cache-aware FastConformer-RNNT model that supports transcription across 40 language-locales and runtime chunk sizes from 80ms through 1120ms.

On Mixpeek, Nemotron 3.5 is useful for agent tools that need low-latency spoken evidence from meetings, calls, streams, and videos. The transcript becomes searchable text, while language tags, timestamps, speakers, and source URIs stay in metadata so the agent can cite the exact evidence instead of returning an ungrounded transcript blob.

Architecture

Cache-aware FastConformer encoder with 24 layers, RNNT decoder, and language-ID prompt conditioning. The cache-aware design reuses encoder context during streaming inference, avoiding redundant overlap computation in chunked ASR.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so nemotron-3.5-asr-streaming-0.6b runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection_id: "col_your_collection",
      documents: [
        {
          document_id: "asset-00412",
          // The model produces text, so it lands in payload. Give the
          // collection a text vector index and embed that text to make it
          // searchable rather than only filterable.
          payload: { extracted_text: modelOutput, source_key: "archive/2026/asset-00412" },
          vectors: { "text-embedding": embeddingOfModelOutput },
        },
      ],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// universal_extractor@v1 runs google/gemini-embedding-2
// (3072-d) over a bucket, with no inference of your own.