What do you like best about VoiceDash?
What made it stick was how little of my setup it asked me to change. VoiceDash sits in the background, I hold a hotkey, talk, release, and the text lands at the cursor in whatever window I happen to be in. Gmail, Slack, Notion, a Google Doc, the prompt box in an AI assistant, a commit message in my editor. There is no separate capture window to open and no copy-paste step at the end. After the first week I stopped noticing the tool at all, which is the highest compliment I can pay to something that runs system-wide.
The AI editing layer is the actual reason to use this instead of what the operating system already gives you. Apple Dictation transcribes what I say. VoiceDash transcribes what I meant. It strips the ums and the false starts, drops punctuation where the sentence structure calls for it, and repairs the mid-sentence restarts that make raw transcription unusable. I speak in long, badly organised sentences when I am thinking out loud, and what arrives in the text field is a readable paragraph rather than a transcript I would have to rewrite from scratch. The gap between those two outputs is the whole product.
Personal dictionary is the feature I set up on day one and have added to steadily since. Client names, product names, the handful of Italian brand terms that every speech engine turns into something else, all of it goes into the list and comes back spelled correctly. If your work involves proper nouns that no general model has ever seen, this is not a nice extra. It is the difference between usable output and a cleanup pass.
The snippet library surprised me. Assign a voice trigger to a block of text, speak the trigger, and the whole block drops in. I keep my email signature, two standard replies I send to inbound partnership requests, and the boilerplate opening for client briefs in there. Multi-line blocks work properly, so these are not just short expansions of a few words. It has quietly absorbed the job I was giving to a separate text expander.
Language handling is stronger than I expected from a young product. I work across Italian and English through the day, sometimes inside the same thread, and the mid-sentence language switcher means I am not opening settings every time the context changes. More than fifty languages are supported and the two I live in are handled well. Accented Italian names come through correctly most of the time, which sounds like a low bar until you have watched other dictation tools fail it repeatedly.
Platform coverage is broad for a tool this size:
- Mac, on both Apple silicon and Intel
- Windows
- Linux, with Debian builds available
- iPhone and Android
The mobile apps are less complete than the desktop builds, but having the same account and the same dictionary on my phone means an idea I capture on a walk ends up in the same place as everything else rather than in a voice memo I never revisit.
The real-time transcript view earns its place because it is where I catch a mangled term before it goes anywhere. I can see what it heard, fix the one word it got wrong, and move on without a second pass later.
Setup took me about five minutes and no training session. Older dictation software wanted you to read passages aloud before it would work properly, and the memory of that put a lot of people off speech input for good. There is none of that here. Install, grant the microphone and accessibility permissions the operating system asks for, pick a hotkey, and start talking. Accuracy on the first sentence was already close to what I get now, months in, which tells you the model is doing the work rather than a profile built up over weeks.
Privacy is handled in a way I can explain to a client without hedging. Processing runs through a zero data retention arrangement, recordings and transcripts are not kept, and nothing I dictate goes into model training. For anyone who dictates client material or anything covered by an NDA, that is the question that decides whether a tool gets installed on a work machine at all. The honest caveat is that this is still a cloud service and audio does leave the device. If your requirement is that nothing ever leaves the machine, you need an on-device tool instead, and no amount of retention policy will substitute for that.
Development pace is what makes me comfortable building a habit around it. There is a public roadmap and feedback board where requests and bugs are tracked in the open, and the team answers reports rather than absorbing them silently. Several complaints I read in reviews from earlier in the year have been addressed since. For a product this young, that responsiveness matters more to me than any single feature on the list.
On price, the free tier gives a thousand words a month, which is enough to work out whether dictation suits your brain and not much more. Pro sits at fifteen dollars a month, twelve if you pay for the year, and lifts the word cap along with the dictionary and snippet features. For daily dictation that is uncomplicated value. Lifetime deals have also turned up on the usual marketplaces, and if one is running while you are evaluating, the calculation changes considerably. Review collected by and hosted on G2.com.
What do you dislike about VoiceDash?
There is real friction here, and most of it lands on Windows.
Hotkey configuration is the restriction I hit first and hardest. On Windows the push-to-talk combination is capped at two keys and has to begin with Ctrl, Shift, or Alt, which collides with shortcuts I already have bound in my editor and my browser. I ended up rebinding one of my own long-standing shortcuts to get out of the way, which is backwards. The Mac side is more flexible, so this reads as a platform gap rather than a deliberate design choice, and the team has said publicly that it is on the list.
Latency is the second thing you notice. Processing happens in the cloud, so there is a pause between releasing the key and the text appearing, and on longer passages that pause is long enough to break the rhythm of thinking out loud. A second or two most of the time, which is fine. On a weak connection or a long stretch of dictation it stretches further. Turning off the audio enhancements in the Windows sound settings shaved something off it in my case. Anyone coming from an on-device model will feel the difference immediately.
Short utterances get clipped. If I hold the key and start speaking in the same instant, the first word sometimes disappears, and with a two or three word sentence that can change the meaning entirely. The workaround I settled on is to start speaking a beat after pressing rather than at the same moment, which fixes it almost completely. Working out what was actually happening took me longer than it should have, because my first assumption was that the transcription was simply wrong.
A few more things worth knowing before you commit:
- The hotkey has stopped responding on me a handful of times, cured by quitting and reopening the app, and it always happens at the least convenient moment
- Dictionary entries are not applied with total consistency, so a term that comes through correctly ten times in a row will occasionally come through wrong on the eleventh
- There is no offline mode of any substance, so on a plane or a bad hotel connection the tool is effectively unavailable
- Custom formatting instructions are followed loosely rather than strictly, which matters if you want output shaped a specific way every time
None of this has pushed me back to typing. It is the kind of friction I expect from a product at this stage, and the items that annoy me most are the ones the team has already acknowledged in public.
The free tier deserves its own note. A thousand words a month is a demonstration rather than a trial. I burned through it inside two days of normal work, which is presumably the point, but anyone who wants a genuine evaluation should plan on paying for a month and treating that as the real test. Review collected by and hosted on G2.com.
