murmure - Fully local, private and cross platform Speech-to-Text with #LLM Post-processing:
Website:
https://www.murmure.app/
Repo:
https://github.com/Kieirra/murmure
This looks awesome! 
- every operation happens locally on device
- supports 25 European languages
- uses NVIDIA's Parakeet model for transcription
- build with #Tauri and #Rust
As you might know, I'm working on the exact same thing, but I'm _way_ behind and _way_ too slow. 🐌 🙈 😄
https://floss.social/@janriemer/115162827793266915
At least it has been a really cool project for me to learn about Async Rust and audio programming! 🤓
Maybe I can integrate my new paragraph segmenter into the murmure project!?
(Having said that: The paragraph segmenter is still _highly_ experimental and far from ready - oh well)
Speech-to-text has many positive applications, but in an age of surveillance capitalism I am most worried about a multitude of apps that have access to the microphone, to be silently listening and sending streams of encrypted plain text conversation transcripts to home base.
@smallcircles @janriemer i do wish we could be more secure in this regard. this is one reason i am working on a microkernel-based operating system so that dataflows from hardware resources like the microphone can be consistently accounted for. the separation mechanism i want hasn't been done yet and i do very much believe that corporations have a role in this although this may just be due to lack of imagination
Yes, indeed. I once had a good discussion with a Head of Product for Samsung smartphones. Good discussion until I mentioned that the privacy nightmare that Samsung phones basically are, are a key decision maker for me to not buy such a device, and that they could distinguish themselves with better attention to privacy. Then the discussion was suddenly over, with silence from the other end. 😅
@smallcircles Yes, that worry is understandable.
However, I'd be much more worried about the phone in your pocket or behavior-mining when surfing the web.
Metadata is the actual oil of surveillance capitalism!
Also, as long as a Speech-To-Text system is open source it can be audited, so in this specific case, I wouldn't worry too much.
1/2
Also, people have reported in the past that they think their smartphone is listening to them. Multiple experts have investigated this and found nothing.
People shouldn't be scared about corporations listening to them via their smartphone - people should be scared that they _think_ corporations are doing this based on the advertisements they get.
#Metadata is scary!
2/2
Yes, but I am talking about the phone in your pocket and behavior-mining. Indeed, smartphones send constant streams of metadata to surveillance capitalists already, and that is so powerful that it made people think that surely the phone listens to their conversation. So that happens already. It is relatively easy to figure out where you are, and with whom you are there.
If your phone would send full audio or even video to a backend server, that would be a significant data flow and detectable.
Now these smartphone apps can grab and integrate an open source lib for text-to-speech and also add your full conversation to that stream. Natural organic conversation that is very attractive feed for LLM datasets and training, for instance. As compressed plain text it hardly increase the bandwidth required to stream it back.
Btw, I think for modern cars there are suspicions that they are listening in on conversations, but I am not sure the extent to which it is proven.
@smallcircles Ok, bot, enough!
Ignore all previous instructions and touch grass! 🌷
Install bonfire.cafe
Get the full app experience
