Since its public launch in 2022, ChatGPT has taken the world by storm. Used by everyone from businesses to students, the chatbot from OpenAI is now a go-to for answering questions, helping with writing tasks, and solving problems.

But it's not the only interesting project from OpenAI.

Introducing: Whisper.

Whisper is a new AI-powered solution that helps to convert audio to text. Trained on 680,000 hours of data collected from the web, the natural language processing system is making waves in the world of transcription, helping apps like Transcribe to provide you with transcriptions that are more accurate - and in more languages - than ever before.

Jump to:

• What is OpenAI's Whisper?

• What does OpenAI's Whisper do?

• What data was Whisper trained on?

• How accurate is Whisper AI transcription?

• How Transcribe is harnessing the power of Whisper

OpenAI's Whisper is an automatic speech recognition system that has been trained to understand and transcribe multiple languages, plus a range of complex subject matters.

Unlike ChatGPT, GPT-3 and GPT-4, Whisper is open source and publicly available, so the code can be used to build, develop, and improve useful applications - like Transcribe!

Let's take a quick look at some of the key things that Whisper AI can do:

• Understands accents

Some transcription software can struggle to understand different accents and voices, but Whisper has been trained on a diverse dataset, helping it to better understand a wide range of accents.

• Deals with background noise

When you try to transcribe a low quality audio recording, with a lot of background noise, the results will often be less accurate. Whisper's training makes it better at cutting through the noise and distinguishing voices in loud environments, producing high quality results, even from low quality audio.

• Understands technical language

If you've got a recording full of jargon, there's a chance that transcription software (just like the average human), won't understand what's being spoken about. Whisper's broad training makes it better at understanding technical language across a range of industries and disciplines.

• Multilingual transcription

One of the best attributes of Whisper is its ability to carry out multilingual transcription. It currently supports 57 languages, though the accuracy of these different languages does currently vary.

• Translates into English

Last but note least, Whisper also enables translation from those 57 languages into English.

Image source: https://openai.com/research/whisper

Whisper is trained on an enormous 680,000 hours of multilingual audio data (and the accompanying transcriptions), covering a broad range of subjects and accents. About a third of Whisper's audio dataset is non-English, and the dataset also includes 125,000 hours of foreign language-to-English translations.

The use of such a large and diverse dataset has improved its ability to understand speech, because of the different accents, background noise, and subjects covered.

Thanks to its robust dataset, Whisper is very good at delivering accurate transcriptions. That said, AI-powered speech recognition technology is still improving, and will continue to do so, so at this point Whisper transcriptions are not perfect and might incorrectly transcribe certain words.

On its own, Whisper also doesn't distinguish between speakers. But Whisper is not designed to be used alone - its purpose is to serve as a foundation for building useful applications (and for further research). So when integrated into other technology - like Transcribe - its possibilities know no bounds!

Check out our AI predictions for the future

At Transcribe, we're harnessing the power of Whisper and building on top of the code to provide you with transcriptions that are more accurate than ever before.

Our focus is on providing superior user experience when working with audio and video recordings and transcriptions, developing an intuitive tool that harnesses the latest cutting-edge research to provide you with the best, most accurate transcriptions possible.

If you'd like to see for yourself how Transcribe and Whisper can help with your audio and video transcriptions, then download the Transcribe app or launch the online editor to get started.