vaibhavs10/incredibly-fast-whisper

Replicate Explore Models
whisper transcription speed-optimization flash-attention replicate transformers

Summary

A highly optimized version of OpenAI's Whisper Large v3 using Transformers, Optimum, and Flash Attention 2, capable of transcribing 150 minutes of audio in under 2 minutes on Replicate.

vaibhavs10 / incredibly-fast-whisper
Original Article
View Cached Full Text

Cached at: 05/08/26, 06:25 AM

# Fast Whisper: High-Speed Transcription on Replicate Source: [https://replicate.com/vaibhavs10/incredibly-fast-whisper](https://replicate.com/vaibhavs10/incredibly-fast-whisper) ## Incredibly Fast Whisper Powered by 🤗*Transformers*,*Optimum*&*flash\-attn* **TL;DR**\- Transcribe 150 minutes of audio in 100 seconds \- with[OpenAI’s Whisper Large v3](https://huggingface.co/openai/whisper-large-v3)\. Blazingly fast transcription is now a reality\!⚡️ Optimisation typeTime to Transcribe \(150 mins of Audio\)Transformers \(`fp32`\)~31 \(*31 min 1 sec*\)Transformers \(`fp16`\+`batching \[24\]`\+`bettertransformer`\)~5 \(*5 min 2 sec*\)**Transformers \(`fp16`\+`batching \[24\]`\+`Flash Attention 2`\)****~2 \(*1 min 38 sec*\)**distil\-whisper \(`fp16`\+`batching \[24\]`\+`bettertransformer`\)~3 \(*3 min 16 sec*\)**distil\-whisper \(`fp16`\+`batching \[24\]`\+`Flash Attention 2`\)****~1 \(*1 min 18 sec*\)**Faster Whisper \(`fp16`\+`beam\_size \[1\]`\)~9\.23 \(*9 min 23 sec*\)Faster Whisper \(`8\-bit`\+`beam\_size \[1\]`\)~8 \(*8 min 15 sec*\) Model createdover 1 year ago

Similar Articles

Introducing Whisper

OpenAI Blog

OpenAI introduces Whisper, an end-to-end encoder-decoder Transformer model trained on large-scale diverse audio data for robust multilingual speech recognition, language identification, and speech-to-English translation. Whisper achieves 50% fewer errors than specialized models on diverse datasets and outperforms supervised benchmarks on speech translation despite not being fine-tuned to specific datasets.

openai/whisper-large-v3

Hugging Face Models Trending

OpenAI released Whisper large-v3, an updated automatic speech recognition model with 128 Mel frequency bins and a new Cantonese token, trained on 5 million hours of data and achieving 10-20% error reduction over large-v2.

rafaelgalle/whisper-diarization-advanced

Replicate Explore

A customizable speech diarization pipeline on Replicate that uses Whisper and Pyannote for fast, multi-speaker transcription with advanced audio preprocessing, noise reduction, and channel separation.