nemotron-speech-streaming-en-0.6b

Name: nemotron-speech-streaming-en-0.6b
Author: nvidia

by nvidia

Ultra-low-latency streaming ASR: 80ms chunks for real-time agent perception

65Kdl/month

0.6Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

nvidia/nemotron-speech-streaming-en-0.6b

Feature URI

mixpeek://transcription@v1/nvidia_nemotron_speech_streaming_v1

Overview

Nemotron Speech Streaming is NVIDIA's cache-aware streaming ASR model built on a FastConformer encoder with RNN-T decoder. It processes audio in configurable chunks down to 80ms, delivering ultra-low-latency transcription for real-time applications where batch models like Whisper are too slow.

The cache-aware architecture maintains state across chunks without re-processing previous audio, making it efficient for continuous streams. It includes automatic punctuation and capitalization. On Mixpeek, it powers real-time audio perception for agents that need to respond to live audio feeds -- call center monitoring, live meeting assistance, and streaming broadcast analysis.

Architecture

FastConformer encoder (cache-aware streaming) with RNN-T decoder. 0.6B parameters. Supports configurable chunk sizes (80ms minimum) with lookahead for accuracy-latency tradeoff. Automatic punctuation and true-casing.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so nemotron-speech-streaming-en-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.