whisper-large-v3-turbo

by openai

Whisper at 216x real-time -- pruned decoder for production-grade ASR speed

4.3Mdl/month

809Mparams

HuggingFace Use in Pipeline

Identifiers

Model ID

openai/whisper-large-v3-turbo

Feature URI

mixpeek://transcription@v1/openai_whisper_large_v3_turbo

Overview

Whisper Large V3 Turbo is OpenAI's speed-optimized variant of Whisper Large V3, achieved by pruning the decoder from 32 layers to 4. This yields a 216x real-time speed factor with less than 1% increase in word error rate compared to the full model.

The model retains the full encoder and multilingual capabilities of Whisper Large V3, supporting 100+ languages. On Mixpeek, it provides the best speed/quality tradeoff for production ASR workloads -- fast enough for batch processing of large video libraries while maintaining near-full accuracy.

Architecture

Encoder-decoder Transformer. Full Whisper Large V3 encoder (32 layers) with pruned decoder (4 layers, down from 32). 809M total parameters. 128 mel spectrogram input. Multilingual, multitask (transcription + translation).

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "my-collection",
  source: { url: "https://example.com/interview.mp4" },
  feature_extractors: [{
    name: "transcription",
    version: "v1",
    params: {
      model_id: "openai/whisper-large-v3-turbo"
    }
  }]
});