Have something to say?

Tell us how we could make the product more useful to you.

-homepage: https://blabby.ai/

Pinned

In Progress

🎉Early Beta Release of BlabbyAI Windows App is Here!

Hey Voice Typers! We’re excited to announce that the early beta version of BlabbyAIWindows app is finally here! Want to Try the Beta Now? Download the Beta Version Your enthusiasm and support have motivated us to bring this to you as soon as possible. However, there’s something important you should know before downloading: ⚠️About the Warning Message The app is now signed with a valid code-signing certificate. However, since it's a new release, Windows may still show a warning message. This is normal—Windows takes some time to fully trust new certificates. If you're comfortable proceeding, you can follow the steps below to bypass the warning. Here’s how to bypass the "unknown publisher" warning: If you encounter any unexpected behavior while using the app, don’t worry! You can easily close the app from the system tray to get things back to normal. We’d love to hear your feedback to make the app even better! Feel free to Request a feature / report a bug in the dedicated feedback portal. https://blabbyai.featurebase.app/ Thanks for being part of this journey! We can’t wait to hear what you think.

BlabbyAI Dev 3 months ago

3

Speech to text prompt library you're welcome to use

Hey! Until you bring out a Linux app, I have to make do with my own very crude Whisper transcription client to fill the gap (there are lots of locally hosted Whisper models but I haven’t found one that uses the API and has direct text input yet). I use this for doing the kind of transcription I do with your tool, but also to capture texts that I then get formatted for use in common situations like creating emails, to-do lists, notes, etc. In the course of making these various prototypes, I've built up a small library of system prompts for what I call text transformation. I'm not sure if this is the official term, but I'm sure that you're familiar with the idea System prompts, which run the dictated text through an LLM to apply some basic transformations I mentioned before that I think it would be really amazing if there was something like a default library in Whisper AI that had the most common of these ready to go so that users could actually not just do direct transcription but also some reformatting or maybe as a secondary functionality. In case you ever do consider moving with the idea, I created an inventory of some of those system prompts earlier this week on GitHub, which you are more than welcome to use for this purpose: https://github.com/danielrosehill/Speech-To-Text-System-Prompt-Library

Daniel Rosehill 18 days ago

In Progress

Anti-hallucination support, pause tolerance

Hey! I'm aware that the underlying engineering in getting all this to work is very complicated and that this would be really hard to achieve, but I said I would throw it out nevertheless. As you helpfully pointed out, ASR speech recognition has the amusing quirk of suffering from hallucinations much as large language models do generally. The manifestation in ASDR being that it will add nonsensical language like “thank you for watching” to the end of a transcription! I've really come to use speech to text as my daily typing method entirely thanks to your great app! So I still feel like I'm discovering how I use it as opposed to just text typing. One thing that I really like to do is to pause for thought while I'm in the middle of dictating. The problem is that if you leave long enough of a gap, you greatly increase the probability of hallucinations and at a certain point they're unavoidable. I don't really have any thoughts on what you could try to do to avoid this. If the pause detection is too aggressive then you run the risk of not capturing user text. But seeing as you've already made this amazing extension, I figure you know a lot more about the engineering than I can speculate about. Perhaps something like a user pause detection threshold setting would be helpful and allow those who dictate cleanly and those who like to pause for thought to choose a setting that best reflects their unique style.

Daniel Rosehill 18 days ago

2

Backup And Export Functionality

Hey! I thought that I had captured this here when I was importing my previous list of suggestions. Forgive me if I'm duplicating. I really like the idea of expanding upon a personal dictionary, although I noticed that when I stop speaking for a while and the tool begins hallucinating, it includes my dictionary words. Any chance you could clarify if the entire dictionary is passed on each current and if so whether that adversely affects token consumption? That aside: My only hesitation about adding lots of interesting modes and uploading and developing a personal dictionary is the same as the one that I have with any other SaaS product, app extension or otherwise: I'm fine with creating lots of data in the cloud so long as I can periodically export it, so just in case you vanish off the face of the earth, I know that I'll have something to fall back on. Perhaps something like a simple CSV or JSON export functionality could be implemented to allow the user to periodically grab a backup copy of their personal dictionary for safekeeping.

Daniel Rosehill About 1 month ago

A feature for post-dictation text reformatting.

There are a number of voice taking apps and productivity tools on the market whose secret sauce is essentially applying a system prompt on top of a dictated text passed through Whisper. From what I've seen, this actually isn't really that hard to do. I use a custom AI frontend and have created dozens of fairly simple system prompts to do everything from converting text into to-do list format through to making it more professional, making it more concise, etc. I know that the current feature set is focused really on dictation and this might be overstepping the boundary into productivity tools, but on the other hand it may actually make sense and save people from requiring multiple components to do this very useful and everyday task. I shared a library of text transformation system prompts on Hugging Face yesterday, which of course are totally open source, and if the idea ever sounds appealing then you are free to use it. https://huggingface.co/datasets/danielrosehill/text-transformation-system-prompts

Daniel Rosehill About 1 month ago

2

The ability to use the app in Chromium and embedded Chrome browsers.

So I am fairly sure that this would be impossible but as usual I thought I would put it out there just as an idea (I guess for those normal people not using Linux, the desktop app will make this irrelevant!): I sometimes use browsers in embedded configurations, things like Electron wrappers or in the current instance Ferdium which is a Workspace OS type tool. Although these are running Chromium under the hood as far as I know, they don't have access to any of the extensions. Hence, while the UIs are very familiar, I greatly miss my favourite productivity tool and have to resort to the awful experience of using a keyboard (how did we ever survive?) Maybe there's some work around or if not hope the Linux app comes to fruition one day!

Daniel Rosehill About 1 month ago

Planned

Microphone level tester

Just being a nuisance again and throwing in a few ideas! From using speech-to-text full-time for a few months now, I'm getting a good handle on what the occasional pitfalls are. One that I've noticed is if your microphone level is accidentally set too low, speech-to-text accuracy naturally declines a lot. Again, as a Linux user, I'm aware that there are peculiarities in this system that most users won't experience. But the one that I've encountered is whatever setting Zoom has that automatically changes the input levels persisting after leaving a Zoom call. Just about any time the Whisper performance seems to be lagging a bit, I check my microphone level and usually it's at something like 50%. I'm guessing that there will be a percentage of people using voice typing for the first time via this extension who will run into all these “rookie errors” and more. I don't know if a decibel meter could be integrated into the extension, but that’s one idea. Another is a warning that might display if the input volume is detected to be below a certain threshold. Something like “you’re too quite! Check your mic levels” (etc)

Daniel Rosehill 2 months ago

1

Support for Deepgram, other STT APIs

Whisper AI is one of those rare technologies that has almost exhausted my ability to generate feature requests because it pretty much just works, which is a great achievement! But in the spirit of sharing ideas and because it's something that I thought about in the early days of testing this stuff out, I wanted to put in the idea that supporting a variety of SGT Cloud APIs might be useful for users. I'm guessing that it would be more hassle than it's worth and it's not really in sync with the Whisper brand (!), but DeepGram's APIs are good, and there are a few others that are useful too. One use case that I would highlight is that I believe a couple of these more niche platforms have explicit support for speaker accents in their API architecture. Users who have very pronounced or more unusual accents might find that the recognition is enhanced with these without having to go through the trouble of fine-tuning a model. Some other providers I’ve come across: Gladia, Speechmatics. As ASR is really heating up, I imagine the list will just keep growing.

Daniel Rosehill 2 months ago

1