granite-speech-4.1-2b-plus

Name: granite-speech-4.1-2b-plus
Author: ibm-granite

by ibm-granite

Speaker-attributed ASR: diarization, word timestamps, and keyword biasing in 2B

95Kdl/month

2Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

ibm-granite/granite-speech-4.1-2b-plus

Feature URI

mixpeek://transcription@v1/ibm_granite_speech_41_2b_plus_v1

Overview

Granite Speech 4.1 2B Plus extends the base Granite Speech model with speaker attribution, word-level timestamp alignment (38.8ms average accuracy), and keyword biasing -- all in a single 2B parameter model. Unlike pipeline approaches that chain separate ASR and diarization models, it produces speaker-labeled, timestamped transcripts in one forward pass.

With a Word Diarization Error Rate (WDER) of 0.9% on the FISHER dataset, it delivers production-grade speaker attribution. Keyword biasing lets you improve recognition of domain-specific terms (product names, technical jargon) without fine-tuning. On Mixpeek, it powers meeting transcription and call analytics pipelines where speaker identity and precise timing matter.

Architecture

Autoregressive encoder-decoder (2B parameters) with multi-task training heads for ASR, speaker attribution, and timestamp alignment. Supports keyword biasing via attention-based shallow fusion. Native vLLM serving support.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so granite-speech-4.1-2b-plus 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.