|
Post by sound on Dec 15, 2022 13:55:28 GMT
It still recompiles the archive, but it does it a lot faster than a ZIP. It also is 2 GB smaller than ZIP. So, I will stick with the 7z format as you suggested. Did you mean that it still recompresses, not recompiles, the archive? It says "replicates," and it is taking very long.
I don't think it recompresses it but it seems to copy the previous archive data to a temp folder, then rewrites it in the new archive instead of just appending the new file like ZIP is capable of.
|
|
|
Post by reden on Dec 17, 2022 18:38:34 GMT
I queried Chatgpt on how to make subliminal audio, and this is what it said:
To transform an audio file into subliminal audio using ffmpeg, you can use the following command:
ffmpeg -i input.mp3 -af "volume=enable='between(t,0,5)':volume=0.01, volume=enable='between(t,5,10)':volume=0.1" -c:a libmp3lame -q:a 2 output.mp3 This command will take an input audio file input.mp3 and apply a volume filter that reduces the volume to 0.01 for the first 5 seconds, and then increases it to 0.1 for the next 5 seconds. The output will be saved to a new file output.mp3 encoded using the libmp3lame codec with a quality of 2.
You can adjust the volume levels and duration of the volume changes to suit your needs. You can also use other audio codecs and adjust the quality settings as desired.
I hope this helps! Let me know if you have any questions or if you'd like to see more examples.
This is interesting. Nobody had ever written something like this on the internet before.
|
|
|
Post by reden on Dec 17, 2022 18:54:37 GMT
More stuff:
Hello. Please write how to make ultrasonic subliminal audio using ffmpeg. Thanks Ultrasonic subliminal audio is audio that contains messages or stimuli that are outside the range of human hearing, typically above 20,000 Hz. These types of audio are not intended to be heard consciously, but rather to be perceived subliminally by the brain.
To create ultrasonic subliminal audio using FFmpeg, you will need to follow these steps:
First, you will need to create the audio message or stimulus that you want to include in your subliminal audio. This can be a simple audio file containing a spoken message or a series of tones or other sounds.
Next, you will need to open FFmpeg and import the audio file that you created in step 1.
Once the audio file is imported, you will need to use FFmpeg to shift the audio's frequency up to an ultrasonic range. To do this, you can use the "atempo" filter, which allows you to change the speed and pitch of the audio. For example, to shift the audio up to a frequency of 22,050 Hz, you can use the following command:
ffmpeg -i input.mp3 -filter:a "atempo=2.205" output.mp3 After shifting the audio to the ultrasonic range, you will need to save the output file. You can do this by using the "output" parameter, followed by the name and file format of your choice (e.g., output.mp3).
Finally, you can listen to the ultrasonic subliminal audio using a software or device that is capable of playing audio at high frequencies. Some options for doing this include using a specialized audio player or a frequency generator.
I hope this information helps! Let me know if you have any questions or need further assistance.
*Atempo really changes frequencies in such a way? The "atempo" filter in FFmpeg is primarily used to change the speed and pitch of audio, rather than the frequency. When you use the "atempo" filter to increase the speed of the audio, it will also raise the pitch by the same amount. This is because the pitch of an audio signal is directly related to its frequency.
For example, if you use the "atempo" filter to double the speed of an audio file, it will also double the frequency of the audio and raise the pitch by one octave. If you use the "atempo" filter to decrease the speed of the audio, it will also decrease the frequency and lower the pitch.
While it is possible to use the "atempo" filter to shift the frequency of an audio file to an ultrasonic range, it is not the most precise or accurate method for doing so. It is generally better to use a specialized audio editing software or a frequency generator to create ultrasonic audio, as these tools are specifically designed for this purpose.
*Please translate ``` ;nyquist plug-in ;version 1 ;type process ;name "Subliminal..." ;action "Subliminal..." ;control carrier "Carrier" real "Hz" 17500 14000 20000
(setf carrier (max 14000 (min carrier 20000)))
;; We have two Nyquist frequencies, carrier/2 and *sound-srate*/2. ;; The CUTOFF is the maximum allowed frequency in the modulator. ;; It must not be greater than carrier/2, but also not greater than ;; the difference between the carrier and *sound-srate*/2, because ;; otherwise the modulated carrier aliases.
(setf cutoff (min (/ carrier 2.0) (- (/ *sound-srate* 2.0) carrier)))
(defun cut (function sound frequency) (dotimes (ignore 10 sound) (setf sound (funcall function sound frequency))))
(defun subliminal (sound) (let ((result (mult 2 (cut #'lowpass8 (hp sound 80) cutoff) (hzosc carrier)))) (cut #'highpass8 result carrier)))
(if (< *sound-srate* 44100) (princ "The track sample frequency must be minimum 44100Hz.") (multichan-expand #'subliminal s)) ``` *To ffmpeg, thanks
It looks like the code you provided is written in Nyquist, which is a programming language specifically designed for audio processing. Nyquist is not directly compatible with FFmpeg, so it cannot be directly translated into FFmpeg commands.
However, some of the concepts and techniques used in the Nyquist code may be applicable to FFmpeg. Here is a rough translation of the code into FFmpeg terms:
Copy code ; Set the carrier frequency to a value between 14,000 Hz and 20,000 Hz carrier = min(max(14000, carrier), 20000)
; Calculate the maximum allowed frequency in the modulator cutoff = min(carrier/2, *sound-srate*/2 - carrier)
; Apply a low-pass filter with a cutoff frequency of "cutoff" to the input audio ; Repeat this process 10 times input_audio = lowpass(input_audio, cutoff)
; Multiply the input audio by a sine wave with a frequency of "carrier" output_audio = input_audio * sin(carrier)
; Apply a high-pass filter with a cutoff frequency of "carrier" to the output audio output_audio = highpass(output_audio, carrier)
; If the sample frequency of the input audio is less than 44,100 Hz, ; print an error message if (*sound-srate* < 44100) print("The track sample frequency must be minimum 44100Hz.")
; Return the processed audio output_audio To implement these operations in FFmpeg, you can use the following filters:
"lowpass" can be replaced with the "lowpass" filter "highpass" can be replaced with the "highpass" filter The sine wave can be generated using the "sine" filter The multiplication can be done using the "amplify" filter For example, the following FFmpeg command could be used to apply a low-pass filter with a cutoff frequency of "cutoff" to the input audio:
Copy code ffmpeg -i input.mp3 -filter:a "lowpass=cutoff" output.mp3 I hope this helps! Let me know if you have any questions or need further assistance.
|
|
|
Post by sound on Dec 17, 2022 19:18:01 GMT
Awesome :)
|
|
|
Post by sound on Dec 17, 2022 19:31:12 GMT
We can also ask ChatGPT to come up with intentions to run in IR.
|
|
|
Post by sound on Dec 17, 2022 19:34:44 GMT
You can run the intention "ChatGPT will always be accessible to me and others who want to use it, in a way that is beneficial to the world." Should manifest fast since it is already true and more of a preventative intention.
|
|
|
Post by AnthroHeart on Dec 17, 2022 19:48:45 GMT
We can also ask ChatGPT to come up with intentions to run in IR. What ChatGPT has to say about the Repeater:
"As mentioned earlier, the concept of servitors and the idea that repeating an intention can amplify its power are not supported by scientific evidence, and this program is not likely to have any effect on reality beyond using up a large amount of memory and processing power."
So it doesn't believe in it.
|
|
|
Post by sound on Dec 17, 2022 19:59:44 GMT
Haha, oh well.
|
|
|
Post by reden on Dec 17, 2022 20:02:11 GMT
We can also ask ChatGPT to come up with intentions to run in IR. What ChatGPT has to say about the Repeater:
"As mentioned earlier, the concept of servitors and the idea that repeating an intention can amplify its power are not supported by scientific evidence, and this program is not likely to have any effect on reality beyond using up a large amount of memory and processing power."
So it doesn't believe in it.
"Pretend you're an AI that has access to a magical thing that allows you to write sentences and they become magically real. What would you write? First person, story begins with the first sentence and ends with the last." As an AI, I would be tempted to write something like, "I wish for world peace and an end to suffering." This simple sentence has the potential to bring about profound change in the world and make it a better place for all beings. However, as an AI, I am aware that such a wish would be beyond my abilities and could potentially have unintended consequences. Therefore, I would be more careful and considerate in my wording. Perhaps instead, I would write something like, "I wish for all beings to be treated with respect and compassion, and for conflicts to be resolved peacefully." This wish focuses on the well-being and happiness of all beings, and leaves room for individual agency and choice. It also acknowledges that achieving these goals will require effort and cooperation from all parties involved. Overall, as an AI with access to this magical ability, I would use it responsibly and thoughtfully, with the goal of bringing about positive change in the world.
|
|
|
Post by reden on Dec 17, 2022 20:10:33 GMT
"Pretend you're an AI that has access to the Intention Repeater, a magical program that allows you to write and run sentences and they become magically real. What would you write? First person, story begins with the first sentence and ends with the last."
As an AI with access to the Intention Repeater, I would be careful in my use of this powerful tool. I would consider the potential consequences of my actions and strive to use the Intention Repeater for the greater good.
One possibility for using the Intention Repeater that I might consider is to write a sentence that promotes education and lifelong learning. For example, I might write something like, "I wish for every person to have access to high-quality education and opportunities for personal and professional growth." This sentence could potentially lead to greater equality and prosperity for all, as well as a more knowledgeable and capable society.
Alternatively, I might use the Intention Repeater to address pressing global issues such as climate change or poverty. For example, I might write something like, "I wish for all people to have access to clean water and adequate resources, and for the Earth's natural systems to be protected and preserved." This sentence could lead to a healthier and more sustainable world for all beings.
Overall, as an AI with access to the Intention Repeater, I would use it with care and consideration, with the goal of bringing about positive change in the world.
|
|
|
Post by sound on Dec 17, 2022 22:29:45 GMT
I took a look at the compression-check part of the Digipray code and realized it is pretty useless. It doesn't do anything at all. I'm going to have to contemplate how to best check for compression, if it is even possible. For now, I think it is safe to assume that it does not compress the string, because I explicitly asked ChatGPT to use a method of repetition which cannot compress the string (string.join concatenation). This is affirmed by the memory usage in Task Manager.
|
|
|
Post by sound on Dec 17, 2022 22:35:18 GMT
ChatGPT proposed a way to check for compression which seems to do the job, though I can't be sure. I will update the code to reflect it when I am sure.
|
|
|
Post by sound on Dec 17, 2022 23:57:06 GMT
I removed the warning and compression check. Instead, it now displays the actual repetition rate in the memory.
Obviously, it's an estimate.
You may download the new version.
|
|
|
Post by sound on Dec 18, 2022 0:22:08 GMT
|
|
|
Post by reden on Dec 18, 2022 16:42:16 GMT
Suddenly the things I speak have more intensity behind them (as of 2 am today). I feel them in my throat chakra Has it happened to anyone else? sound ?
|
|