FYI, I made a new post for the audio analysis Emily Echo’s Audio Investigation that has information from the audio file in excel format so people can manipulate it as they see fit.
2 Likes
I wonder if we could close this thread? If that’s possible, I’m not sure LoL. But might make it easier.
Well to be honest I don’t think there is… Said that, I didn’t check my self… could do, what if 
haha. I must say to, that any bits (here I talk about the bit, as 8 bits in a byte) information encoded in the audio is irrelevant if you re encode it in a different format. (like event in wav) if the video itself is a copy of the stream it will change the bit inside.
What I mean is, let say the video is legit, and the same as the stream. The audio must be extracted as is, not re encoded in whichever format. Using ffmpeg: ffmpeg -i video.avi -vn -acodec copy audio.aac. But the extension of the audio must correspond to the one in the video. You’ll know with this cmd: ffmpeg -i video.avi it will be generaly the last entry (stream 1:0 …)
Once you have the exact sound file, you cannot do any modification on it, otherwise it will change the bits inside. Even augmenting the gain will change it all.
Now if the information (if it has) in hiding in the wave form itself (as in code morse) then having the conversion in wav is legit. And even augmenting the gain too.
Just be aware that fiddling with embedded bit information is tricky as a little manipulation can change the bit number.
2 Likes
Hey! A little more info, we were running audio thru an RTTY software which expects to be receiving a signal from a shortwave radio. The program won’t run if it does not see an external audio input. So, we ran it into the in line jack from a phone and tricked the program into thinking it was hooked to a radio…
2 Likes
Hmm… Just read the wiki on RTTY and as I read it, I was wondering how it sound. And the more I read about it, the more I doubt it sound like the video. Then at the end, it has an sound exemple 
File:RTTY.ogg - Wikipedia
It really don’t sound like the video 
Check out the knowledge center on this forum. It can sound differently. Good article there.
1 Like
I assume this is measured with a 45 baud rate, correct?
I have not looked into pure Baudot much or line code and I am therefor not sure how you came to your resulting byte sequence. I can only assume this being correct, although the audio wave pattern does not look like a block pulse with a regular clock pulse, or symbol rate. Maybe something to do with the shift, but my knowledge on this currently fails me. Maybe worth a topic in the #knowledge-center
I did look into the more common RTTY, which this certainly is not.
Info: KLOTH.NET - RTTY signals and other URLs linked in previous posts.
For common RTTY the high/low frequencies commonly used sure do not match. Listening to various examples, clearly indicates this audio not being common RTTY.
2 Likes
See update further down this topic here!
I have been working on another method, because there certainly is a clear pattern with the silences and non-silences. They have distinct lengths, with the silence parts varying between two lengths only and the non-silenced parts having 9 different lengths. Below some of the work I did using Sonic Visualiser and Gimp.
Note: Grid size = 16px. I numbered each non-silenced audio part according to the table below. The longer silence is numbered 0, while I did not number the silence between each other non-silenced part of audio.
Nr Grid blocks Pixels
9 5 80
8 4.5 72
7 4 64
6 3.5 56
5 3 48
4 2.5 40
3 2 32
2 1.5 24
1 1 16
0 black silence
Counts:
1 (55), 2 (105), 3 (105), 4 (59), 5 (13), 6 (4), 7 (4), 8 (2), 9 (2) > Total of 349 bits of sound
Longer silence/cut: 0 (39)
Short silence/cut: 312
Note: Above image shows the results. Requires download and zooming in real close for proper visibility, due to its width.
Number sequence result
Note: Number between parenthesis is total amount of non-silence audio parts
0
744239132522312341231334 (24)
0
444 (3)
0
5123342561 (10)
0
243421 (6)
0
42211132243111 (14)
0
1315322462 (10)
0
33423342114323315322345 (23)
0
433222332132212 (15)
0
1334234143422334133313222424323 (31)
0
8322342331 (10)
0
4 (1)
0
322235 (6)
0
233 (3)
0
23143523323 (11)
0
2231223 (7)
0
32 (2)
0
13134132134244122 (17)
0
3233132 (7)
0
2333314343324 (13)
0
42 (2)
0
3 (1)
0
25111271422 (12)
0
4 (1)
0
223312 (6)
0
327221243422 (12)
0
133 (3)
0
252142 (6)
0
46442 (5)
0
44173422 (8)
0
5 (1)
0
13413415322442123233 (20)
0
324 (3)
0
124322432 (9)
0
413 (3)
0
123413233133322223331 (21)
0
3222494265138 (13)
0
2332222 (7)
0
1 (1)
0
433 (3)
No larger silence follows.
Not yet sure what to do with this, but feel free to use and speculate about.
2 Likes
wow, nice work all, back to the grind for me to see what can be found.
I used WaveForm software and saw there were 4 channels placed on top of each other.
Did any one else see that?
Which usually means it is enrypted.
2 Likes
You can open audio files with FLDIGI. File > Audio > Playback. It can’t read MP3s though, so you need to convert to .wav first. But there really doesn’t seem to be anything.
1 Like
Ok, so I tried a different approach, but I think someone who knows WTF they are doing needs to do this right.
I noticed the silent part after all the morse code sounding part was never higher than 5khz, while the part with went up to 20khz in places. I thought that was funny, and kept thinking of carrier frequencies, etc.
So I applied a filter and removed all audio under 5khz, and was left with this file:
Then, I could see audio up there, couldn’t here it, so I shifted the frequencies down to the 100hz-200hz (voice) range.
Well, there was something there, but faint, so I had to amp it up to 11 so I could hear it. That’s when I got this file. It’s 1am here, and I’ve been at it a while, but I swear I can hear Emily saying ‘Bork’ over and over.
So, that’s my cue that I need to let some professionals see if there’s anything here, and hopefully get more than a Bork out of it. Also, I should go to bed now. 'Night!
5 Likes
I have reached out to the ham radio group that Emily has marked as an interest in her Facebook page, still waiting to be accepted into the group so i can ask questions, I also messages one of the members in the mean time but no response yet, they will prob think I’m kookoo lol
3 Likes
What exactly do you mean with 4 channels? The audio directly stripped from the Twitch video is stereo, which means 2 channels (left and right). Care to explain?
I’ve gone through all the binary data sets that were posted by me, ewrk and Difficultlevel, and attempted Baudot and Baudot-Murray (ITA1 and ITA2) decoding on them. No meaningful looking results.
I’m still pondering how to perform some meaningful decoding on the data from the analysis by DevilinPixy and co, based on the lengths of the silent parts of the audio.
I was wondering if the silences were purely random, and the discrete lengths were just because the output of the random number generator had been multiplied by some constant - in C you would do something like this:
int silence_length = (rand() % 11) * SOME_CONSTANT;
However I plotted a distribution graph of the values Retro_Ally posted in the other thread, and they don’t look random.
There are only two lengths of silence, the regular short silence between each short piece of audio from the loop and then the somewhat longer silence at irregular intervals. The latter could certainly be close to a randomly generated interval when looking at the amount of pieces of noise in-between. You would basically be looking at the numbers I put within parenthesis in my ‘Number sequence result’ a couple of replies back. Of course this somewhat longer silence would replace a short silence at that spot. The only things that really changes length is the short pieces of audio themselves (9 variants), which appear less likely to be randomly generated. The shorter variants appear more often than the longer ones:
1 (55), 2 (105), 3 (105), 4 (59), 5 (13), 6 (4), 7 (4), 8 (2), 9 (2) > Total of 349 bits of sound
Some other things I noticed:
- At the spots where silences are, they are cut straight into the audio of the loop, as if audio data has just been erased without affecting the audio waves at all. You can see clear straight up/down lines in the audio waves at beginning/end of each bit of sound. Can be compared with another loop adjacent to it and they are identical, besides the cuts made.
- The full stream contains a repeating loop of audio, exactly 53 times.
- The first 22 loops are at a slightly lower volume.
- After about 47m53s, the volume goes up slightly to repeat another sequence of 31 loops
- The distortion, or cutting into the audio loop starts at about 1h24m53s
4 Likes
The blanks areas are not actually blank. There is still audio information in there, but the volume has been enormously reduced. You can isolate the blank areas, amplify them, and put the sound back. The volume reduction increases towards the centre of the blank, suggesting a roll-off filter has been used.
1 Like
Ok Ive been eccentric by the HAM radio group, have submitted a question, waiting on a response
3 Likes
You are correct, they are not blanks where no audio data exists at all, it is indeed just dead zero volume. I could have probably worded that better. However, isn’t that required for an audio file, as time continues, or is it actually possible to have true gaps of nothing in-between?
I am not seeing roll-off towards the center of the silence, but an instant drop of the volume to zero at the start of a silence and then right at the end of the silence, straight back up/down again to continue the audio that was supposed to be there. See below example:
Ok my post got removed and flagged as spam!! How rude!! I guess that’s not going to work then, I am at a loss lol
It’s not a criticism, it’s an observation. Something that might help us understand what this thing is.