Skip to content

Whisper large-v3-turbo

Fast speech recognition and translation across 99 languages

Visit official site

What Whisper large-v3-turbo is

The model is described as an automatic speech recognition and speech translation model from OpenAI, trained on more than 5 million hours of labelled data. It generalises to many datasets and domains in a zero-shot setting.

What you can do with it

  • Recognise and translate speech in 99 languages
  • Generalise to many datasets and domains without fine-tuning
  • Process long audio files chunk by chunk
  • Predict the language of the source audio automatically

Getting started

  • Install the Transformers, Datasets and Accelerate libraries
  • Load AutoModelForSpeechSeq2Seq and AutoProcessor from the Hugging Face Hub
  • Build an automatic speech recognition pipeline with the pipeline class
  • Load and prepare audio from a Hub sample dataset or a local file

Model notes

  • The model is a fine-tuned version of Whisper large-v3
  • Decoder layers are cut from 32 to 4 to raise speed
  • The reduction may slightly lower quality

Pros & cons

✓ What we like

  • Decoder layers reduced from 32 to 4 for faster running
  • Compatible with all Whisper decoding strategies

! What to watch out for

  • Fewer decoder layers may slightly lower quality
  • Speed gain comes from the fine-tuned decoder cut

FAQ

How many languages are supported?

The feature list names 99 languages for speech recognition and translation.

What changed from Whisper large-v3?

The decoder layers were reduced from 32 to 4 to raise speed.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review