I tried Google’s AI dictation app, but it couldn’t replace Wispr Flow

Great language models have given AI the advantage it has today, but one of its more obscure uses is enhanced voice dictation. We saw a version on Gboard, which lets you talk to your phone in natural language and convert your voice to text, complete with emoji and punctuation. I rely on a standalone tool called Wispr Flow to achieve similar results on my Mac and avoid writing long text messages and emails.

I always thought that since Google is already so good with voice dictation on Android via Gboard, a Google desktop tool would wipe the floor with anything else on the market. But it turns out I was completely wrong. Eloquent AI Edge from Google It works in the same way as Wispr Flow, with the only exception that it works much worse, to the point that it is practically unusable for any kind of serious work.

Google Eloquent advantage of AI is an offline voice dictation tool that works without the Internet and uses Gemma, Google’s class of language models that run entirely on the device. On your first installation, you will be asked to download a basic offline model supporting 2 billion parameters. If your laptop has more than 16 GB of RAM, you can even download a 12 billion parameter model for better performance. Since my Mac falls into the first category, I ran my tests using the 2 billion parameter model. Edge Eloquent is available on iOS and macOS, but not Android, which is understandable since Gboard’s advanced voice dictation feature already works.

Did you know? It took Edge Eloquent at least 15 tries to write “frustrating” in the title above. And that’s the kind of hassle I have to deal with every time I use it.

This Google tool has a built-in polishing feature that cleans up the text before pasting it into the text box. More often than I’d like, it changes the very structure of a sentence, sometimes altering tense, omitting words, and rephrasing it into perfect AI-written copy rather than something I’d write in my own language. And when I turn off the polishing feature to prevent exactly this, Edge Eloquent keeps all the ums and ahs and just pastes everything as is.

In any case, I have to spend a considerable amount of time going back over everything sentence by sentence and editing what I just said. Unfortunately, there’s no middle ground: if there was a slider to choose the amount of polish, I’d be a much happier Google AI Edge Eloquent user.

What further contributed to my editing workload was that Edge Eloquent often ignored my commands for new lines or scripts much more frequently than Wispr Flow. Secondly, I found the offline Gemma model to be pretty mind-blowing. There have been at least a couple of cases where he added completely unrelated phrases to the text that I hadn’t even spoken.

What I mentioned above are the problems that keep interrupting my workflow so frequently that I can’t rely on it for any serious work, like writing this story. However, if I enunciate my words more clearly and maintain a slow dictation pace (although I’m okay with a little more editing along the way), I can get AI Edge Eloquent to work at a satisfactory level. It feels even more amazing when I realize that their Mac app is doing all this magic completely offline without stressing the system too much.

I kept an eye on my Mac’s activity monitor while dictating a piece, and Edge Eloquent only showed brief spikes in CPU, GPU, and RAM usage, none of which seriously affected any of the other apps open on my system. I’m sure the peaks would be noticeably different if I had installed the 12 billion parameter model, but so far, everything seems well controlled. And the fact that it’s completely free to use is a big selling point in its favor.

Since Wispr Flow works in the cloud, it has an obvious advantage. I’ve had far fewer structural problems with it, although it’s not completely bulletproof. It has its fair share of issues, although far fewer compared to the Google AI Edge Eloquent. For example, it almost always gets brand names right, something that Google’s alternative often fails to do, and I can maintain a faster dictation pace with Wispr Flow without needing to emphasize every word to get it right. That’s a particularly important trait for a writer like me to focus on when working on a story.

Although Wispr Flow It is a paid product and requires a constant internet connection to work, I would prefer it over Google’s AI Edge Eloquent simply because of its greater reliability. When I focus on writing, having to frequently go back and edit what the dictation tool did wrong not only takes up a lot of time, but also breaks my flow. I just can’t afford to lose my train of thought as often as I do with Google’s product. Maybe Edge Eloquent works better on a more robust system running the 12 billion parameter model, but what I have is not enough for me.

That doesn’t mean I’ll ditch Edge Eloquent entirely. Since it is almost 60-70% in terms of transcription and polishing, I can still use it in situations where Wispr Flow fails. For example, it will be useful when I’m on a train with a patchy network and I need to quickly change a story before reaching my destination. Wispr Flow recently suffered an outage resulting in slower responses and, in many cases, no transcription at all. The nice thing about an on-premises model is that it is completely unaffected by cloud outages.

Google has done a fantastic job of finding a specific use case for their LLMs on devices. Considering how efficient it is on hardware while running completely offline, it works decently well. Since it’s completely free, it could be perfectly suitable for personal use, whether that’s writing emails from time to time or writing prompts for ChatGPT. But for more professional use, Wispr Flow remains the undisputed king for me, while AI Edge Eloquent will be my fallback tool.