Whisper

Whisper is a multilingual speech recognition encoder–decoder (seq2seq) model family from OpenAI, available across sizes from tiny to large-v3 / turbo variants on the Hub.

Adapted for this playbook from the 🤗 Transformers documentation by Hugging Face. Official page: https://huggingface.co/docs/transformers/model_doc/whisper. Images © Hugging Face (documentation-images) unless noted. This is not a substitute for the upstream docs — verify against the current version.

Covers Hugging Face pages: model_doc/whisper

Executive summary

Use Whisper via Pipeline for transcription prototypes, then harden with batching, VAD and domain fine-tuning for production call-centre or media workflows.

Quick usage

from transformers import pipeline

asr = pipeline(
    task="automatic-speech-recognition",
    model="openai/whisper-large-v3",
)
print(asr("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac"))

from transformers import AutoProcessor, WhisperForConditionalGeneration

processor = AutoProcessor.from_pretrained("openai/whisper-large-v3-turbo")
model = WhisperForConditionalGeneration.from_pretrained(
    "openai/whisper-large-v3-turbo", device_map="auto"
)

Tips

• Match audio sample rate to the processor (typically 16 kHz)
• Use timestamps parameters for meeting notes
• See Audio tasks

Official API

Whisper model doc.

What to do next

ViT.