Review

• Author: @richardanderson001
• Repo: openai/whisper
• Title: Well built with easy install

Summary

Whisper is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio and is also a multitasking model that can perform multilingual speech recognition, speech translation, and language identification.

Most heavily developed recently by:

• https://github.com/jongwook: technical staff at OpenAI

pros

• great examples on colab,
• very easy install process, was able build in a conda env with python=3.8
• active development: 41.4k stars, 368 watching, 4.7k forks

cons

• heavily developed by one person, support is likely subject to their bandwidth constraints
• does not have issues enabled
  • because of this errors seem to have migrated to discussions (eg. PyPI build out of date, installation issues, etc.)
  • reports of issues hosting on AWS lambda layer, with model being too big

Next

Would love to get a better perspective on what the biggest issues for this repo are and if anyone is using it in production.

The Show and tell has helpful stories.

Also examples exist on https://huggingface.co/spaces/aadnk/whisper-webui