Rendered at 09:53:39 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
crazygringo 17 hours ago [-]
I think we're about to see a dictation revolution.
I've been coding with Claude, and I pretty much do it exclusively by voice now, simply because it's so much faster. And I kind of feel like Scotty from Star Trek IV, when he holds up a Mac mouse to try to talk into it.
I've been using VoiceInk [1] which looks like it's basically the same as this, but has been around for longer.
What has really made it work for me is using a Bluetooth media control like [2], where I use Karabiner Elements to remap its play/pause button to the dictation keyboard shortcut. I map rewind to option-backspace to delete the last word, fast-forward to shift-enter to insert a newline, a lower button to enter to submit my prompt, and the volume up/down buttons to scroll up/down. I've also just ordered to Xiaomi Bluetooth remote [3] that includes a microphone itself to see if I can get it to work by speaking directly into it and using it as the microphone -- there are a couple of open source projects to turn it into a Mac microphone directly. Since I'd like to be able to talk more quietly instead of into my Mac.
But what I'm REALLY waiting for is the ability to use one button for dictation, and a second button for issuing commands for a local LLM to interpret. So I hold down the voice button which transcribes "I think we need to catch that" and then the second button and go "change catch to cache, like c-a-c-h-e". Or hold down the second button and go "switch to VS code". I don't want something as finicky as macOS Voice Control, I want a local LLM I can speak naturally to.
I can very much see a future where I spend the majority of my "work" time using a Apple TV-type remote.
I am having great results by mapping the press to dictate onto my pedal.I have been vibecoding without using my hands for the past four months with great results.It's amazing using the computer without touching it.I can go on for hours with a couple of pedals.I use them to switch windows, dictate and press enter.That's really all you need.
vardalab 15 hours ago [-]
The one thing about the Bluetooth microphones is that there is this annoying lag that it's very hard to get used to, even when using AirPods which you think would be optimized for this. It's just untenable in my opinion because I could never get used to the 3-400 millisecond delay that I had to deal with. I ended up using simple Apple wired headphones as my go-to solution eventually. They actually produce a least a modest strain on ears when being in for hours on end, at least in my situation. I do like your idea about the Bluetooth multimedia remote controller.
13 hours ago [-]
kartik017 15 hours ago [-]
i have fixed this problem in betterwispr btw ;)
pegasus 15 hours ago [-]
You fixed bluetooth lag??
kartik017 3 hours ago [-]
yup kinda of
pegasus 21 minutes ago [-]
Strange that one would troll their own Show HN...
Grombobulous 17 hours ago [-]
I can see the appeal from a workflow perspective, though from a “social” perspective I don’t see myself enjoying talking to myself all day.
Bonus points if you work in an office. This kind of workflow would be a nightmare.
…unless it means we get our private offices back.
dnautics 16 hours ago [-]
I have a friend who has reporposed a stenographer's mask
jasonm23 7 hours ago [-]
I hear ~~the Harkonnen~~ Amazon are hiring.
redanddead 2 hours ago [-]
ok THAT'S crazy lol
ProblemExplorer 5 hours ago [-]
> I can very much see a future where I spend the majority of my "work" time using a Apple TV-type remote.
Why don't you use an AI agent to code up an app for your phone that does this and more? If a phone is too heavy or bulky, what about an ESP32?
shostack 10 hours ago [-]
I have been wanting a Bluetooth ring myself so I can just tap my finger and have it trigger my airpods which would initiate live voice or dictation on my phone or computer.
I have not considered input for editing functions but that makes a lot of sense. Maybe a swipe gesture or something with a little capacitive pad on the ring.
cjonas 17 hours ago [-]
i forked handy and built 2 modes. One for dictation and one for quick agent actions that uses the pi harness with local models and jev style classifiers so it can read screens and interact with elements (via accessiblity tree, screenshots, os scripts, bash, mcp, etc). Its nice because in agent mode you can just give it instructions on how to respond (like dictation that can read the context of the current page or input). I've been pretty happy with it so far and debating if i should release it. Just not sure if the world needs yet another vibe coded agent system.
heliostatic 10 hours ago [-]
Please do! It sounds like a useful fork
patrickk 14 hours ago [-]
Great ideas here thanks! I’m going to see if I can whip up a Windows equivalent, using Autohokey instead of Karabiner.
ulimn 11 hours ago [-]
How do you keep your prompts from getting just ramblings as you do the thinking?
It sounds like a fun way to "code", but I just can't imagine how it works in reality.
crazygringo 10 hours ago [-]
So this is the actual amazing thing. Claude does absolutely fantastic with rambling. As I am dictating I may change my mind three times as I keep imagining new cases or new constraints. And then at the end of my prompt I will say something like OK this has been a whole lot of brainstorming in different directions. I'd like you to summarize from me what you interpret as the final decisions and the reasoning behind it and the reasoning for why we are not going to adopt the proposed alternatives. And then tell me if you think I'm missing anything.
And it does. I don't even have to write out my final thoughts neatly summarized myself. Claude figures it out on its own and saves me a bunch of time. It's much faster for me to read it's logically laid out conclusions and confirm that that is actually what I had meant. Never once have I seen it get it wrong, but quite frequently it raises additional points for me to consider or highlights potential inconsistencies in what I have said.
So the actual session itself with Claude includes a whole lot of rambling and brainstorming, but then the actual documented artifacts of the session are the nice clean summary of what we figured out. And then, for example, when we move from a brainstorming design phase into a coding implementation phase, I'll start a brand new session so that it's context isn't polluted by the brainstorming and rambling but is rather starting off with the nicely organized document that had been created as the artifact from the previous session.
And this is why dictation works so well as a feature. Because it means I frequently wind up essentially "oversharing" my product requirements with Claude, and it will frequently suggest potential solutions that I had overlooked because I was already assuming that the solution would take a particular form. So it becomes surprisingly collaborative rather than me just dumping requirements into Claude.
anon373839 9 hours ago [-]
Qwen 3.8 Flash Next also works very well with this, and frankly even the much smaller 3.6 35B-A3B did, too.
estearum 10 hours ago [-]
That's actually why this method is so powerful: ramblings are actually good as prompts. Seriously, try it.
sunsetMurk 11 hours ago [-]
I actually like this about talking vs. typing. often I feel like I get better results since I think give it more context around spec and decisions.
Razengan 17 hours ago [-]
I'm glad we're getting to a point where voice can be the main UI, and that some people are actively using it to their benefit, but I hope it doesn't become THE primary interface, even with AIs.
I find speaking tiresome, somehow, and if I had to talk all the time to my computer with text input being a fidgety "accessibility" option, I'd retire to a monastery and just become a monk.
Agree with the need for different buttons for different voice contexts though.
Though I think it will become a dedicated AI key on some keyboards.. How about the Right Alt/Option on MacBooks? Does anybody use that? It could be the "AI Anywhere" input key, optionally defaulting to voice, and while F5 could continue to serve as a literal dictation key..
pmoriarty 15 hours ago [-]
Speech also has a lack of privacy, and it can annoy people nearby.
Imagine everyone on public transport speaking in to their phones to navigate and control them.
A microphone sensitive to subvocalization could maybe get around some of these problems, but it'd still be kind of weird.
dnautics 15 hours ago [-]
I upload documents to a remarkable, mark it up, and have Claude build changes there.
Big fan of Handy. It's simple, offers all the models with a great picker and the author is involved in building a general STT model runtime. It does streaming flawlessly with Nemotron, which mind-bogglingly many apps don't offer.
I checked out the alternatives mentioned in this thread and they appear more over-loaded, "sloppy" and commercial. Handy just does the thing without extras.
I recently switched to Voiceink which also does similar. I'm trying to find the edge this has over that but can't find it
abyssin 18 hours ago [-]
VoiceInk took a little time compiling but it's been working flawlessly for the three people that use it now for dictating consultation notes.
alexgoodhart 16 hours ago [-]
lmao
patrickk 15 hours ago [-]
Can you link to the exact website? Multiple similarly named projects show up.
evulhotdog 15 hours ago [-]
I recently switched from Handy to Fluidvoice, and like it much better. The UI is much better and offers a better and more comprehensive feature set in comparison.
kartik017 15 hours ago [-]
i do liked fluidvoice, can't compete with them honestly but do give it a try
nxpnsv 19 hours ago [-]
They both use parakeet so I guess they would perform similar.
pylotlight 18 hours ago [-]
ye parakeet redux has been my new fav, that and tdt v2 both working well for my own clone app that was made.
ramraj07 19 hours ago [-]
How good is parakeet (assuming it’s Sota oss) compared to wisprflow? Wisprflow is magical ime
rcarmo 17 hours ago [-]
As much as I like seeing this, the space is really crowded, and I think the real value is not in single dictation for input, but in two other things: diarization (essential for call recordings, and a staple in video calling services, but hardly touched for in-room meetings and brainstorming) and the next step, which is cleaning it all up for actually useful notes. Not just the transcript, but better variations on meeting notes that you can customize, link to previous existing information, etc.
polyterative 16 hours ago [-]
In my own vibe coded implementation during the recordings the live transcript gets sent for validation and allows me to take better decision while in the call itself.I'm using a a small local model for processing.Also like when it detects bullshit told by others in real times so I can immediately reply.Fun
prash20026 17 hours ago [-]
Isn't that what Google just released although it is mac only for the moment.
I've been using https://talat.app/ for this, I've not been using it for long, a few weeks, but it's working well so far.
eajr 16 hours ago [-]
There are a number of free apps that do this now: ghostwhisper and anarlog are the first two that come to mind. It is not hard to vibe your own too. There are open source models that do diarization.
rcarmo 2 hours ago [-]
As someone who works in the field… having models that do something is not the same as having a workable solution.
kartik017 16 hours ago [-]
yup this is what i have been focusing on.
you can try the notetaking feature which i have build
Royce-CMR 17 hours ago [-]
What’s really shocking is this and many of the competitors are all using OpenAI whisper - which came out in 2022. (Once upon a time, OpenAI really did open research)
It’s surprising in 2026 this is still the default “best” all rounder. Kudos to the early OpenAI team and the open community efforts / improved models since then.
kartik017 15 hours ago [-]
no it's a really old and bad stt model i am using parakeet v3 here which is way superior
akurilin 10 hours ago [-]
Superwhisper trains their own S1 models for this, you can use either a smaller local version of them or a beefier one in the cloud
sunnybeetroot 16 hours ago [-]
Is there anything better for speech to text model?
eajr 16 hours ago [-]
Yes, parakeet. Ive switched all my apps to using it. Fantastic model
sunnybeetroot 16 hours ago [-]
That is great! Just no fast inference like cerebras or groq support it
vardalab 16 hours ago [-]
Parakeet is blindingly fast on modern hardware. My MacBook is just so fast, it's not even funny. But despite that, I still often use OpenWhisper because it just does better for accented voices. I also fine-tuned Qwen ASR on my voice, and that one is faster than Open Whisper, but I think I might even go back to Open Whisper eventually Because it still makes a lot of weird mistakes And I think it is over-indexed towards the fine tune And substitutes quite often wrong dictionary words for ambiguous words.
kartik017 15 hours ago [-]
openwhispr dx is really bad, i made this product because i was tired that no one is working on accuracy and dx in general.
betterwispr main goal is to have accuracy and best dx possible.
sunnybeetroot 16 hours ago [-]
Server side though?
kartik017 16 hours ago [-]
parakeet v3 betterwispr uses that.
shantnutiwari 54 minutes ago [-]
These seem to be coming once a week now. And most suck, I dont know why people keep creating new ones?
I wrote a blog about this[1] "Audio transcription is worse in 2026 than it was in 2016"
What a whiny blog, just press the dictation button on your keyboard to have direct dictation capabilities without any rewriting. If you don't want this AI-based rewriting, just don't use it.
scosman 18 hours ago [-]
If folks like local-AI dictation, but want it optimized for meetings recordings (like Granola/Otter/Notion) check out Biscotti. Free, local, separates voices, voice identification, AI summaries, etc.
Otter was really like Hey you, subscribing? Great! Two recordings per month.
d4rkp4ttern 10 hours ago [-]
On MacOS, dictation was solved a while back with Parakeet V3, and I haven't seen anything that has a better speed/accuracy tradeoff when talking to AIs, where slight inaccuracies are ok since they can "read between the lines".
I got annoyed by various UI and other things in tools like VoiceInk, Handy etc and Claude-Coded my own CLI dictation tool called VoxType that uses parakeet-mlx:
This is just an example of where we are today with software today: people building their own tools, and not bothering to market it (or perhaps I'm contracting myself here ;).
cootsnuck 5 hours ago [-]
Spot on. My backup local notetaker app just stopped releasing updates and became broken so I got annoyed and just whipped up my own to not lose access to my old recordings and also to just use going forward.
So much better to just add my own features and slowly customize it to suit my needs.
I have a fun feature where I can retranscribe with multiple local models and then compare transcripts and optionally label/correct them to accumulate data on which models are most accurate over time.
And if MacOS 26 does it right - all of this is no longer necessary for us to vibe fork/ code on weekends :-)
P.S. Handy looks slick. May be able to ditch mine
arach 6 hours ago [-]
voxd.cc/minivox - tiniest
talkie.to - bells and whistles
esafak 17 hours ago [-]
MacOS 27 still has room for improvement; there is a measurable starting latency and error rate.
zalthor 11 hours ago [-]
I was so excited when they presented "improved dictation" during WWDC, and people reviewing the developer previews seem to think it was a stand-in replacement for things like Whispr Flow, but I am little disappointed with the eventual release. It's not a disaster, but I feel like there's so many small little things they could add to meaningfully improve it. Things like adding a dictionary of words that you can maintain for proper nouns or choosing which mic to use for dictation.
CharlesW 15 hours ago [-]
For me (M1 Studio) it works at least as well and as quickly as Handy or GhostPepper, which are considered two of the best. (Turn on with Keyboard > Dictation, works anywhere you can type text.)
Been using Handy I am pretty happy but good to see more options, specially OSS.
kartik017 16 hours ago [-]
handy is bloat and really bad dx in general.
this product is just 14mb binary.
bluehatbrit 13 hours ago [-]
I understand you made BetterWhispr and Handy would be a competitor, but if you're going to dunk on them it would be good to add some real detail and reasoning. It's okay if you just didn't get on with it, but this is HN where we want to hear about design decisions, trade-offs, and the motivation for building this over using one of the many alternatives.
I've not used either and I'm sure they're both great, but just saying an alternative is bad isn't particularly useful to anyone here.
kartik017 3 hours ago [-]
i used handy a year ago, i think it was my first voice dictation app trial.
the reason it is bad, it's a 150-200mb binary i remembered downloading, it was not accurate for my accent, was missing lot of words, i tried wisprflow and it was accurate, the dx was top notch and i was a annual plan subscriber, but i think they are cost cutting on inference and all of the features which i mentioned got severely degraded.
this was the main motivation to build betterwispr, again this is my personal opinion, i was just thinking out loud. really sorry that people get offended in general.
apparent 4 hours ago [-]
Interested to give this a try. I've tried Wispr Flow and found it to be quite good, but a bit pricey for my taste. I guess I assumed there were equally good models that were cheaper, and I suppose this is one of them.
I'll be curious to see how the existence / popularity of tools like this end up affecting the pricing of paid tools like Wispr Flow.
zalthor 11 hours ago [-]
Interestingly enough, I was working on something very similar this past few of days[1]! A couple of things that I did differently -
One, in the post-transcription cleanup phase, instead of programmatically fixing the text, I just feed the text with a prompt + dictionary of words into an LLM. Initially tried doing this with the on-device model from Apple in macOS 27, but that was too slow. I've now been playing around with gpt-oss-20b on Groq, which appears to be fast and effective. Although I'm curious how well your programmatic approach works, I might try some sort of a bake off over the next few days. But in the long run, I think I might still stick with an LLM-based formatter because you can have far more interesting formatting considerations or formatting prompts based on the app that you are using or the style or tone of your dictated text.
Two - When building the dictionary of words to fix, I'm also including a recording of the word and having the transcription model just predict draft out variants to use as replacements.
The native / built-in API SpeechAnalyzer (introduced last generation on iOS and MacOS) works quite well.
Are folks finding it lacking / needing alternatives?
throw03172019 18 hours ago [-]
Voice shortcuts etc
kartik017 16 hours ago [-]
[dead]
sajithdilshan 13 hours ago [-]
Even though this looks really good I don’t see myself using this.
Mostly because I could use this only when I’m working from home, and even then after a few minutes I’ll get tired of speaking to myself and I actually type way faster than I speak.
kartik017 3 hours ago [-]
i will be really honest with you i use to think the same, typing is slow when you are in a flow state, i have 80 wpm speed, but since i started using voice dictation apps my productivity increased, clarity of thought and everything in general made more sense.
GodelNumbering 15 hours ago [-]
Just tried it out as I was looking for something that does this exact thing well (macbook). So far I dictated about 5 different things in different contexts, it worked without a single error! Good choice to pick parakeet v3.
Nice work, thanks for building it!
kartik017 15 hours ago [-]
ooh thanks so much man, if you face any issues do open a thread on github.
Curious how this handles Chinese, or mixed Chinese/English mid-sentence.
Parakeet's 25 languages look European-only, so I assume Whisper is the only CJK path here?
kartik017 3 hours ago [-]
yeah sadly you will need to use whisper model there.
galkk 6 hours ago [-]
I’m typing much faster than I’m speaking, but I’d like to control my os with voice more. Hope we’ll get there eventually.
arach 6 hours ago [-]
what type of control would you like? For me the key is to give my agents more useful OS abstractions and I do this via https://lattices.dev - this gives my projects a nice semantic representation and action space for my OS
I don't do everything. Also fully expect Siri and AppIntents will continue to improve
having said that, what would you like to control?
galkk 2 hours ago [-]
Funny enough - navigate between apps, targeted control of where I want to land in particular app/window etc.
I’m a dinosaur and still read the code that my ai has written and do a pass with my own eyes of other people PRs. Saying and landing precisely in things like “open this PR in VSCode”. Does it require many actions? No. Would it reduce friction? Hell yea. Doing all that while the ide (example, again) will suddenly decide to rebuild the indexes or run extension updates etc is breaking the flow. Being able to jump into fully loaded and prepared place that is ready to go - it would be very good.
I will admit. First it started with the need to use like 2 ides in everyday work (there was a time at Google when IDEA was defunded and eclipse was sole supported Java ide, but I hated eclipse wholeheartedly and used it only for debugging), and now with ai coding I really feel how my fluency with ides is degrading day after day. I remember myself even 5 years ago (and more so 10-15) - I, somehow, much more rely on mouse or things like “find action/command palette” than I did in past.
frenchie4111 18 hours ago [-]
This is awesome. I really want to get something like this integrated directly into my agent IDE (https://getness.dev).
The main thing I need is some way of auto-sending a "end of message" button (like a newline would do) at the end of the message so that it auto-sends the chat.
Not sure the exact right way to implement that, but it seems like something that could be general purpose. Maybe a setting per-active app?
HnUser12 17 hours ago [-]
I use handy.computer in push-to-talk mode. It inserts text in the active text box when you let go. It also supports auto-submitting at the end, so maybe that would help? I think it can send new line.
rcarmo 17 hours ago [-]
Many people are using Nemotron (NVIDIA provides Linux binaries for ARM and Intel that you can shove into a server and do audio over websockets for them).
regularfry 18 hours ago [-]
Push-to-talk?
frenchie4111 17 hours ago [-]
Yes I want push to talk in ness
But trying to implement it directly into ness hasn't worked that well. It seems like betterwispr/wisprflow/etc all seem to belong better as their own app, and we just need good integration points between the push-to-talk app and the text consuming apps (like the IDE)
tamilini92 13 hours ago [-]
[flagged]
jayshah5696 11 hours ago [-]
1 feature I want is using Llm local or cloud for cleaning. I like open whisper and typewispr for both.
Playing with anything that just works. Also allow for cleaning existing text. (Written not by speech) In some way.
sunsetMurk 11 hours ago [-]
[dead]
gekoxyz 18 hours ago [-]
I expected dictation to be much more useful on the computer, but since I can type decently fast I find myself using it far less than I expected.
crazygringo 17 hours ago [-]
I can type crazy fast, ever since I was a kid.
Nevertheless I dictate almost exclusively with Claude Code now, because it's such better ergonomics. And it's still significantly faster.
But a big benefit is that I'm never hunching, I'm never straining my wrists, none of that stuff. Dictation is just so much better for your neck and back.
gekoxyz 12 hours ago [-]
Yeah Claude Code is one example where I use dictation most of the time (unless I'm in the office). Even if I take back some things that I say there it can be useful context for the LLM to parse and understand what I'm doing.
The thing is that for long emails, long text messages, or in general when I have to think about what I want to communicate and I have to write decently, ASR isn't helping me as much as I thought it would. Maybe I'm using it in the wrong way, and I could build a better pipeline if I added LLM postprocessing to my text, but I don't have enough memory on my computer for this kind of task yet.
vardalab 16 hours ago [-]
Yeah, this is a very good point. I just lean back and speak. I find it much better. You can actually think now better because you can actually focus on the problem, admittedly, I never bothered to learn how to type fast because I was too lazy And my fingers are beefy. And I also like it when I'm literally just looking at something and giving a feedback to model while I'm speaking and it is being transcribed and being acted upon. I mean, this is all is kind of like science fiction. I'm surprised that there's so much negativity about all this AI or as we officially now refer to as Si.
11 hours ago [-]
jakobov 16 hours ago [-]
This is cool. I also built a dictation app that has advantage of using three state-of-the-art models in parallel. Which increases accuracy and reduces p95 significantly. You can check it out at https://zwhispr.com/
copperx 6 hours ago [-]
Are those three models better than just using Elevenlabs Scribe V2?
alexgoodhart 16 hours ago [-]
2000 words free per month.
jakobov 15 hours ago [-]
ya tokens cost money. Unfortunately there is a limit to how much I can subsidize.
handfuloflight 12 hours ago [-]
What about adding a wake word detection so I can just say "Hey wakeWord" to activate the STT? Then maybe detect silence to finish. Completely hands free?
vivzkestrel 4 hours ago [-]
- real time speech to voice clone that works with obs studio when?
- use case: record gaming videos in your voice and in real time get cloned voice back
- why? too much noise in surroundings
- whose gonna build this one?
kartik017 3 hours ago [-]
this is a cool usecase will add it in backlog
Benard-dev 17 hours ago [-]
How are you handling local transcription? We opted for an offline-first for FetchMark archiver for same privacy reason and it doubled our retention.
Congrats on shipping!
kartik017 16 hours ago [-]
[dead]
epaga 18 hours ago [-]
I'm a bit confused, isn't "Wispr" a trademarked term? Why would you pick a name that could easily be shut down?
kartik017 16 hours ago [-]
nope it isn't trademarked actually
handfuloflight 15 hours ago [-]
U.S. trademark rights are conferred just by use in the market, not necessarily requiring an application.
Na6z 18 hours ago [-]
This is pretty neat, I built something similar a couple months ago. I'd say try to support Windows or Linux next.
This is my entry into the world of vibe coded Wispr Flow clones. Runs locally on the Whisper AI speech recognition model file - the small variety - just under 200mb.
I thought the small model won't be any good, but it's actually great and barely causes the GPU any usage.
Like other people in this thread mentioned, I switched pretty much exclusively to working with voice. Whether it's responding to text or talking to Claude or writing an email.
It's funny. I've auditioned about 30 clones but none does exactly what I want. Utter.to is the closest but it's buggy because it's mostly vibecoded. I need to vibecode my own.
honkycat 16 hours ago [-]
I've been trying to figure out how to code and build while on the bike /treadmill at the gym.
Want a remote terminal into my Mac that I can voice control.
kartik017 16 hours ago [-]
yeah you can do try this out, it's really smooth and has good native macos ux in general.
it's just a 14 mb file so no bloat as well.
ProofHouse 18 hours ago [-]
To be honest, I used Wspr so much I only recently started realizing just HOW bad it gets trying to 'improve' your diction, especially coding. I have seen totally opperate directives. I had no idea it was potentially occuring, but it gets past a point where it is WAY to confident in guessing what you meant, sometimes to the extent of changing DON'T to DO, for example.
kartik017 15 hours ago [-]
agree with your point, but wisprflow has been degraded recently and i wanted a solution which is free and works basically
null-phnix 17 hours ago [-]
[flagged]
stevyhacker 14 hours ago [-]
[dead]
kartik017 22 hours ago [-]
Hey HN,
I’m building BetterWispr, a free and open-source voice dictation app for macOS.
I wanted to make voice typing more accessible without requiring a monthly subscription or sending every recording to a cloud server.
You can hold Option + Space, speak naturally, and have the transcription inserted directly into whatever app you’re using.
A few things I’ve been working on:
- Local speech recognition using Whisper, Parakeet, or Apple Speech
- Automatic cleanup of filler words and repeated phrases
- Custom vocabulary for names and technical terms
- Different writing styles depending on the app
- Optional meeting transcription and summaries
The project is open source under Apache 2.0.
I’m still improving the experience, especially around transcription accuracy, speed, and reliability across different Macs.
I’d love feedback from people who regularly use voice dictation.
What would make you switch from your current dictation tool to an open-source alternative?
I've been coding with Claude, and I pretty much do it exclusively by voice now, simply because it's so much faster. And I kind of feel like Scotty from Star Trek IV, when he holds up a Mac mouse to try to talk into it.
I've been using VoiceInk [1] which looks like it's basically the same as this, but has been around for longer.
What has really made it work for me is using a Bluetooth media control like [2], where I use Karabiner Elements to remap its play/pause button to the dictation keyboard shortcut. I map rewind to option-backspace to delete the last word, fast-forward to shift-enter to insert a newline, a lower button to enter to submit my prompt, and the volume up/down buttons to scroll up/down. I've also just ordered to Xiaomi Bluetooth remote [3] that includes a microphone itself to see if I can get it to work by speaking directly into it and using it as the microphone -- there are a couple of open source projects to turn it into a Mac microphone directly. Since I'd like to be able to talk more quietly instead of into my Mac.
But what I'm REALLY waiting for is the ability to use one button for dictation, and a second button for issuing commands for a local LLM to interpret. So I hold down the voice button which transcribes "I think we need to catch that" and then the second button and go "change catch to cache, like c-a-c-h-e". Or hold down the second button and go "switch to VS code". I don't want something as finicky as macOS Voice Control, I want a local LLM I can speak naturally to.
I can very much see a future where I spend the majority of my "work" time using a Apple TV-type remote.
[1] https://github.com/Beingpax/VoiceInk
[2] https://www.amazon.com/Satechi-Bluetooth-Multimedia-Remote-C...
[3] https://www.notebookcheck.net/Xiaomi-Bluetooth-Remote-2-Pro-...
Bonus points if you work in an office. This kind of workflow would be a nightmare.
…unless it means we get our private offices back.
Why don't you use an AI agent to code up an app for your phone that does this and more? If a phone is too heavy or bulky, what about an ESP32?
I have not considered input for editing functions but that makes a lot of sense. Maybe a swipe gesture or something with a little capacitive pad on the ring.
It sounds like a fun way to "code", but I just can't imagine how it works in reality.
And it does. I don't even have to write out my final thoughts neatly summarized myself. Claude figures it out on its own and saves me a bunch of time. It's much faster for me to read it's logically laid out conclusions and confirm that that is actually what I had meant. Never once have I seen it get it wrong, but quite frequently it raises additional points for me to consider or highlights potential inconsistencies in what I have said.
So the actual session itself with Claude includes a whole lot of rambling and brainstorming, but then the actual documented artifacts of the session are the nice clean summary of what we figured out. And then, for example, when we move from a brainstorming design phase into a coding implementation phase, I'll start a brand new session so that it's context isn't polluted by the brainstorming and rambling but is rather starting off with the nicely organized document that had been created as the artifact from the previous session.
And this is why dictation works so well as a feature. Because it means I frequently wind up essentially "oversharing" my product requirements with Claude, and it will frequently suggest potential solutions that I had overlooked because I was already assuming that the solution would take a particular form. So it becomes surprisingly collaborative rather than me just dumping requirements into Claude.
I find speaking tiresome, somehow, and if I had to talk all the time to my computer with text input being a fidgety "accessibility" option, I'd retire to a monastery and just become a monk.
Agree with the need for different buttons for different voice contexts though.
Though I think it will become a dedicated AI key on some keyboards.. How about the Right Alt/Option on MacBooks? Does anybody use that? It could be the "AI Anywhere" input key, optionally defaulting to voice, and while F5 could continue to serve as a literal dictation key..
Imagine everyone on public transport speaking in to their phones to navigate and control them.
A microphone sensitive to subvocalization could maybe get around some of these problems, but it'd still be kind of weird.
I checked out the alternatives mentioned in this thread and they appear more over-loaded, "sloppy" and commercial. Handy just does the thing without extras.
https://developers.google.com/edge/foresight
you can try the notetaking feature which i have build
It’s surprising in 2026 this is still the default “best” all rounder. Kudos to the early OpenAI team and the open community efforts / improved models since then.
betterwispr main goal is to have accuracy and best dx possible.
I wrote a blog about this[1] "Audio transcription is worse in 2026 than it was in 2016"
1 https://write.as/shantnu/audio-transcription-is-worse-in-202...
https://github.com/scosman/Biscotti
I got annoyed by various UI and other things in tools like VoiceInk, Handy etc and Claude-Coded my own CLI dictation tool called VoxType that uses parakeet-mlx:
https://pchalasani.github.io/claude-code-tools/tools/voxtype...
Install this skill for your coding agent and let it help you set it up, there are many config options:
Or if you must do it yourself, This is just an example of where we are today with software today: people building their own tools, and not bothering to market it (or perhaps I'm contracting myself here ;).So much better to just add my own features and slowly customize it to suit my needs.
I have a fun feature where I can retranscribe with multiple local models and then compare transcripts and optionally label/correct them to accumulate data on which models are most accurate over time.
And if MacOS 26 does it right - all of this is no longer necessary for us to vibe fork/ code on weekends :-)
P.S. Handy looks slick. May be able to ditch mine
this product is just 14mb binary.
I've not used either and I'm sure they're both great, but just saying an alternative is bad isn't particularly useful to anyone here.
the reason it is bad, it's a 150-200mb binary i remembered downloading, it was not accurate for my accent, was missing lot of words, i tried wisprflow and it was accurate, the dx was top notch and i was a annual plan subscriber, but i think they are cost cutting on inference and all of the features which i mentioned got severely degraded.
this was the main motivation to build betterwispr, again this is my personal opinion, i was just thinking out loud. really sorry that people get offended in general.
I'll be curious to see how the existence / popularity of tools like this end up affecting the pricing of paid tools like Wispr Flow.
One, in the post-transcription cleanup phase, instead of programmatically fixing the text, I just feed the text with a prompt + dictionary of words into an LLM. Initially tried doing this with the on-device model from Apple in macOS 27, but that was too slow. I've now been playing around with gpt-oss-20b on Groq, which appears to be fast and effective. Although I'm curious how well your programmatic approach works, I might try some sort of a bake off over the next few days. But in the long run, I think I might still stick with an LLM-based formatter because you can have far more interesting formatting considerations or formatting prompts based on the app that you are using or the style or tone of your dictated text.
Two - When building the dictionary of words to fix, I'm also including a recording of the word and having the transcription model just predict draft out variants to use as replacements.
[1] https://github.com/goose42/TalkToMyMac
Are folks finding it lacking / needing alternatives?
Mostly because I could use this only when I’m working from home, and even then after a few minutes I’ll get tired of speaking to myself and I actually type way faster than I speak.
Nice work, thanks for building it!
https://github.com/opennookorg/betterwispr
also do star it helps.
I don't do everything. Also fully expect Siri and AppIntents will continue to improve
having said that, what would you like to control?
I’m a dinosaur and still read the code that my ai has written and do a pass with my own eyes of other people PRs. Saying and landing precisely in things like “open this PR in VSCode”. Does it require many actions? No. Would it reduce friction? Hell yea. Doing all that while the ide (example, again) will suddenly decide to rebuild the indexes or run extension updates etc is breaking the flow. Being able to jump into fully loaded and prepared place that is ready to go - it would be very good.
I will admit. First it started with the need to use like 2 ides in everyday work (there was a time at Google when IDEA was defunded and eclipse was sole supported Java ide, but I hated eclipse wholeheartedly and used it only for debugging), and now with ai coding I really feel how my fluency with ides is degrading day after day. I remember myself even 5 years ago (and more so 10-15) - I, somehow, much more rely on mouse or things like “find action/command palette” than I did in past.
The main thing I need is some way of auto-sending a "end of message" button (like a newline would do) at the end of the message so that it auto-sends the chat.
Not sure the exact right way to implement that, but it seems like something that could be general purpose. Maybe a setting per-active app?
But trying to implement it directly into ness hasn't worked that well. It seems like betterwispr/wisprflow/etc all seem to belong better as their own app, and we just need good integration points between the push-to-talk app and the text consuming apps (like the IDE)
Playing with anything that just works. Also allow for cleaning existing text. (Written not by speech) In some way.
Nevertheless I dictate almost exclusively with Claude Code now, because it's such better ergonomics. And it's still significantly faster.
But a big benefit is that I'm never hunching, I'm never straining my wrists, none of that stuff. Dictation is just so much better for your neck and back.
- use case: record gaming videos in your voice and in real time get cloned voice back
- why? too much noise in surroundings
- whose gonna build this one?
Congrats on shipping!
do star it on gh if you liked it!
I thought the small model won't be any good, but it's actually great and barely causes the GPU any usage.
Like other people in this thread mentioned, I switched pretty much exclusively to working with voice. Whether it's responding to text or talking to Claude or writing an email.
https://imgur.com/YJyz05r
Want a remote terminal into my Mac that I can voice control.
it's just a 14 mb file so no bloat as well.
I’m building BetterWispr, a free and open-source voice dictation app for macOS.
I wanted to make voice typing more accessible without requiring a monthly subscription or sending every recording to a cloud server.
You can hold Option + Space, speak naturally, and have the transcription inserted directly into whatever app you’re using.
A few things I’ve been working on:
- Local speech recognition using Whisper, Parakeet, or Apple Speech - Automatic cleanup of filler words and repeated phrases - Custom vocabulary for names and technical terms - Different writing styles depending on the app - Optional meeting transcription and summaries
The project is open source under Apache 2.0.
I’m still improving the experience, especially around transcription accuracy, speed, and reliability across different Macs.
I’d love feedback from people who regularly use voice dictation.
What would make you switch from your current dictation tool to an open-source alternative?
Website: https://betterwispr.com/