Welcome!

By registering with us, you'll be able to discuss, share and private message with other members of our community.

SignUp Now!

Sounds, signals and software: A Crootn deepdive

Hoppo

Elite
Founder
Joined
Jan 12, 2021
Messages
896
Very interesting (to me at least).

One idea that was enlightening to me about voice/pressure waves was:

Let’s say a couple people in a kitchen are talking, and there is a big of chips on the table. You want to understand what they are saying, but you are 500 yards away with no microphone. If you can get a high res / high speed camera on that bag of chips, you can convert very tiny movements on that bag detected by the camera into speech.

Combine those ideas with how many cameras and microphones are everywhere, how easy it is to convert “pressure waves” to text, you can basically figure out anything said anywhere and store it. The big challenge now is there is just so much data to process, but as the tools progress no one will have any secrets that aren’t known on a raid somewhere unless they are very careful

I mean that’s entirely feasible technology. I wonder if it would be better to construct a best guess waveform out of the data rather than attempting to transcribe directly. Fidelity of audio is the biggest confound to voice recognition, and those sorts of issues would be greatly magnified in constructed audio. But I bet you could build an audio file that a human could understand. The human brain is so many orders of magnitude better at such recognition.
 

shiv

John
Administrator
Founder
Member
Joined
Dec 1, 2020
Messages
14,169
Hot damn kid. That's some serious shit.
I’ve done some cool stuff in matlab with financial data and stonks too. Maybe a different thread on a different day.

You can do web scrapers really easy on there
 

shiv

John
Administrator
Founder
Member
Joined
Dec 1, 2020
Messages
14,169
I mean that’s entirely feasible technology. I wonder if it would be better to construct a best guess waveform out of the data rather than attempting to transcribe directly. Fidelity of audio is the biggest confound to voice recognition, and those sorts of issues would be greatly magnified in constructed audio. But I bet you could build an audio file that a human could understand. The human brain is so many orders of magnitude better at such recognition.
See my sound cloud upload earlier in the thread. I did twinkle twinkle little star in matlab.
 

shiv

John
Administrator
Founder
Member
Joined
Dec 1, 2020
Messages
14,169
That’s impressive. I’ve never worked with Matlab - are you able to write out raw byte data with it I gather?
Matlab is kinda like Python. It’s a compiled language - good for scripts.

I just write the equation for a sin wave and then digitize it (maybe 10,000 samples in a second). For each note I just set the f in the equation to match the frequency of the given note.

y = sin(2*pi*f*t)

At this point in the discussion we are just a hop skip and a jump away from the “Voice of God” weapon or the MIT tool that connects to the back of the head and lets you “think” a word and the device will say it.

Before words become pressure waves, they are just electrical signals in our brains. And the electrical signal for each word is unique
 

Latest posts

Top Bottom