DEV Community

Cover image for How to run Whisper on Google Colaboratory
Kazuki Yonemoto
Kazuki Yonemoto

Posted on • Updated on

How to run Whisper on Google Colaboratory

Whisper is a general-purpose speech recognition model open-sourced by OpenAI.

According to the official article, the automatic speech recognition system is trained on 680,000 hours of multilingual and multitask supervised data collected from the web.

πŸ“– Introducing Whisper

I was surprised by Whisper’s high accuracy and ease of use.
Whisper provides so useful command line that you can feel free to try it.

πŸ“– Command-line usage

I'll present how to run Whisper on Google Colaboratory here.
You can refer to this Colab notebook if you want to try Whisper immediately on Google Colaboratory.

πŸ“– Colaboratory whisper-mock-en

Create a new Colab notebook

You need to crate a new Colab notebook from your Google Drive at the first.

Colab notebook

You have to make sure your notebook is using a GPU. To do that, change a runtime type to GPU from the menu.

GPU from the menu

Install package

You need to install a package like the following line to run Whisper.

# Install packages
!pip install git+
Enter fullscreen mode Exit fullscreen mode

Add folders

Add this code to create new folders when you click the play button.

import os

# Add folders
checkContentFolder = os.path.exists("content")
checkDownLoadFolder = os.path.exists("download")
if not checkContentFolder:
if not checkDownLoadFolder:
Enter fullscreen mode Exit fullscreen mode

Upload an audio file

You have to upload at least an audio file into the content folder after you've installed packages and added folders.

Transcription with Python

You can modify the file name of the audio and target language to translate.

import whisper

fileName = "sample.m4a"
lang = "en"
model = whisper.load_model("base")

# Load audio
audio = whisper.load_audio(f"content/{fileName}")
audio = whisper.pad_or_trim(audio)

mel = whisper.log_mel_spectrogram(audio).to(model.device)

# Output the recognized text
options = whisper.DecodingOptions(language=lang, without_timestamps=True)
result = whisper.decode(model, mel, options)

# Write into a text file
with open(f"download/{fileName}.txt", "w") as f:
  f.write(f"β–Ό Transcription of {fileName}\n")
Enter fullscreen mode Exit fullscreen mode

Download a transcription file

It would be easy to download a transcription file if you added this code.

from google.colab import files
!zip -r download"")
Enter fullscreen mode Exit fullscreen mode

Let's run Whisper on Google Colaboratory

Let's check how your code works on Colab notebook.
I prepared a Colab notebook to use Whisper, so you can copy this notebook.

πŸ“– Colaboratory whisper-mock-en


I hope this tutorial got you to generate your first transcription.

Top comments (1)

Sloan, the sloth mascot
Comment deleted