Back to Blog

Introducing Voice Writer: AI-powered dictation and grammar correction

Mar 22, 2024

Writing is hard. Sometimes, you have the perfect idea in your head, but getting it into words can feel like an uphill battle. That’s writer’s block—it takes time and effort to transform messy thoughts into polished, coherent writing. And all the time spent fixing grammar and formatting, and the whole process feels like a chore.

That’s why I’m excited to introduce Voice Writer: a tool that changes the way you write. With Voice Writer, you simply speak your thoughts—this is the fastest way to capture your ideas. Then, let AI powered by GPT-4 refine your text, ensuring it’s clear, polished, and in the right style.

How does Voice Writer work?

Voice Writer is built to make writing faster, easier, and better. Here’s how it works:

  1. Speech-to-Text: Using state-of-the-art speech recognition models, Voice Writer captures your words with exceptional accuracy.

  2. AI-Powered Grammar Correction: Powered by GPT-4, Voice Writer fixes grammar and punctuation, even if you repeat words, pause mid-sentence, or use informal phrasing. It polishes your text automatically, turning it into professional-quality writing.

Why existing dictation tools don't work

Voice dictation isn’t new. Tools like Google Docs, Microsoft Word, and Apple Dictate have offered it for years. However, I’ve always struggled to produce good writing with them. The results often require extensive editing, which defeats the purpose of using dictation software in the first place.

Here’s an example of a short note I dictated using Microsoft Word:

LLMs nowadays are very versatile. Zay. Can be combined. Together with each other to do tasks that they weren't originally designed for. For example for speech processing, one of the most popular. Models for speech recognition is the open AI with models. And they? However, needs to be. Fine tune with. Another. IOM, like Harlem model can be combined together and fine-tuned end to end.

As you can see, it’s barely usable. While most of the words are recognized, there are too many issues:

  • Punctuation: Sentences are fragmented and poorly punctuated, making it hard to follow.

  • Speech Recognition Errors: Words like "they" are misrecognized as "zay," and "Llama" becomes "Harlem." (my Chinese accent is not helping)

  • Lack of Fluency: Since I’m forming my thoughts as I speak, the output is messy and disjointed.

Editing something like this takes significant time and effort - making this barely faster than typing it out normally.

Now, compare that to the output from Voice Writer:

LLMs nowadays are very versatile; they can often be combined with other models to extend their capabilities beyond what they were originally trained to do. An example of this is speech recognition. One of the most popular models is OpenAI's Whisper model - this model, by itself, only performs speech recognition, but it can be combined end-to-end with a large language model like Llama and fine-tuned for enhanced performance.

This output requires minimal editing to be usable in a blog post or email message. With Voice Writer, I save hours of editing time, allowing me to focus on crafting my ideas instead.

More ways to write with voice

I use Voice Writer for a variety of tasks every day, making writing more productive and fun:

  • Emails and Slack Messages: Never stress about how to word an awkward email to your boss or overthink that quick Slack message to a teammate.

    Documents: Speak your thoughts even if it's a dense and technical design doc or report. Voice Writer will help you sound polished and professional.

    Blog Posts and Book Reviews: I write a lot of blog posts and notes on various topics, now what used to be a chore takes me minutes.

Excited to launch and see how others integrate it into their workflows! We have lots of improvements planned, from enhanced features to even more user-friendly updates.

Try Voice Writer for free today!