Summary

A Google Colab notebook is introduced for transcribing and translating audio files using OpenAI's Whisper model.

Abstract

The article discusses the author's fascination with Spanish writer Enrique Vila-Matas and the challenge of accessing untranslated interviews due to language barriers. Upon the release of OpenAI's Whisper, a general-purpose speech recognition model, the author creates a Google Colab notebook for transcribing and translating audio files. The notebook, named "Transcribe and Translate with OpenAI's Whisper," is tested with an Enrique Vila-Matas mini-doc, resulting in a translated .TXT file. The author reflects on the implications of AI-generated content and the abundance of materials, quoting Vila-Matas' fear of being "finished" in a world of endless text generation.

Bullet points

The author is a fan of Spanish writer Enrique Vila-Matas but struggles to understand untranslated interviews.
OpenAI releases Whisper, a general-purpose speech recognition model.
The author creates a Google Colab notebook, "Transcribe and Translate with OpenAI's Whisper," for transcribing and translating audio files.
The notebook is tested with an Enrique Vila-Matas mini-doc, resulting in a translated .TXT file.
The author discusses the implications of AI-generated content and quotes Vila-Matas' fear of being "finished" in a world of endless text generation.

How to transcribe and translate with OpenAI’s Whisper

Use Whisper for free with this Google Colab notebook

One of my favorite writers is Enrique Vila-Matas, a Spanish writer obsessed with books and the reading life.

I’ve read everything he wrote that I can find in English, but too many of his interviews are untranslated, trapped in scholarly Spanish prose too intricate for my feeble Spanish skills to decipher.

Whisper + Google Colab

When Open At released Whisper this week, I thought I could use the neural network’s tools to transcribe a Spanish audio interview with Vila-Matas and translate it into English.

I began by looking for Google Colab notebooks I could use to tap into this open-source general-purpose speech recognition model.

I couldn’t find one, so I wrote my own Colab notebook that anybody can use: “Transcribe and Translate with OpenAI’s Whisper.”

A screenshot of my Google Colab notebook.

My coding skills are primitive at best, but I managed to bend all my knowledge into a few paragraphs.

After a bit of work, I had built an AI translation machine. With my Google Colab account, I could use this general-purpose speech recognition model to both transcribe or translate audio files.

I tested my Colab notebook with audio from an Enrique Vila-Matas mini-doc I’ve always wanted to watch, “Cafe Con Shandy.”

The whole process was magical and cumbersome, but now I have an enormous .TXT file containing a pretty solid translation of a once indecipherable bit of video.

AI and “the abundance of materials”

That chunk of newly translated text took some extra reading to sort out, but I finally understood what they were talking about in the video.

I even found a sentence in there that’s applicable to our world where GPT-3 can generate endless text and Whisper can translate it.

“Of course, and it is a fear that I think the writer has, it is the fear of being finished,” Vila-Matas notes in the video that I translated with the help of Whisper.

“This idea of finishing sometimes has to do with your writing, with a different question or that would seem to oppose the conclusion — which is precisely the excess, the abundance of materials.”

We now live in a world of AI-generated abundance, and writers will need to find a place in this sea of content.

For me today, that means something new and interesting to read. For me tomorrow, that means an existential struggle to find readers in a sea of AI-generated stories and images.

Vila-Matas is right. The writer will never be finished. But it is a little scary and thrilling to see our excessive future, overwhelmed by our “abundance of materials.”