[Abandoned] Wav to snr converter

[Kubik]

As announced in our 11th community update, we decided to open source wav_to_snr (repository link). This is a very simple application that allows you to convert audio files to a format that is compatible with BattleForge called SNR.

SNR is compressed a lot, and it is lossy compression (over 70% of data is lost). Sample count is the same as WAV, but for most samples only 4 bits instead of 16 bits is used for the value, that means 16 possibilities instead of 65536 possibilities. It is not that bad as it might look at first look. There is some function (difference from previous 2 samples) applied so the dynamic range does not need to be that big, and only precision is lost.

Currently it works this way: for each block of 32 consequent samples it tries all "configurations" and calculates all samples in the block, then pick the one that have on average smallest difference from the original sound. The problem with this is that if the original audio is decreasing at the end of block, but the compressed one is increasing it, it can still be the one with smallest difference, and if next block start lower than the source, the difference between the last sample of the block and first sample of the next block will be even greater because of that.

In theory "configuration" should be picked so that whole audio wave is as similar shape as possible, instead of absolute difference for each sample averaged over independent blocks. But comparing shapes of waves is much more complicated thing, requires much more computation time, and what is most important it is much more complicated to write.

So if anyone will improve the algorithm it would be great 🙂 right now it is kind of unusable for voice 😞 but for some simple effects it sounds kind of maybe acceptable.

If you want to just play existing file check out:

[Kaliber84]

Sounds like a job for a Fourier transform. I will try to give it a look in the upcoming weeks once I get access to my Laptop that runs Labview again. I am not too familiar with audio formats and their compression methods but I've worked with image compression and proprietary file structures already, so I might be able to help out.

What programming language are you using? Not too comfortable with my programming skills but I'll make do.

Edited September 25, 2021 by Kaliber84

[Kubik]

I am using Rust,and If I would buy LabView it would be my second most expensive software by far. For other audio files I think you can find enough documentation easily. Snr is different story thou 😞 there are linked resources in the repo, and the repo is trying to reproduce `EA XAS ADPCM` codec that is used by other BF files. We also know that EALayer3 (EL31 and I think L32P) are supported by BF. EALayer3 is mp3 with some bit changes (no I can not say what exactly those changes are 😢). https://bitbucket.org/Zenchreal/ealayer3/src/master/ is code that can for some mp3 files convert them to valid snr files, we recently discovered the (almost complete) set of conditions that the file needs to met to be correctly converted, and even thou mp3 claims compression of 75-95% it actually sounds better than our attempts to reproduce `EA XAS ADPCM` codec.

If you will have any questions about the formats, or any other question feel free to reach to me (preferably on discord, but forum also works 🙂 ).

[Kaliber84]

@Kubik Thanks for the info. I wasn't suggesting to buy Labview, but via my university I get a free license which I am using. It is easier for me to use than to dive into Rust which I am not familiar with. I'll give an update once I have any kinds of results (or questions 😛).

[Kubik]

I did not say you suggested it 🙂 I just checked if they have some free version and found how expensive even the most basic version is 😞 so I can not help with the LabView part 😞

[Majora]

Hey everyone,

As mentioned in the community update, we ran into issues implementing the voicelines. We would love to get the help from the community figuring this out. 

Issues we ran into:
• The game sometimes ends a new voiceline with a loud snap. This seems to be unrelated to what voiceline is being played, and seems to happen about 20% of the time. 
• The game sometimes completely crashes after playing a new voiceline, most of the time repeating the last word of a line a couple of times before completely shutting down Skylords Reborn. 
• Wounded and Strike voicelines are not referenced in the ACK file, but the game does have audio files (and references them correctly) for existing units. We are not sure how to add these specific voicelines into the game. 

We have currently not been able to find a logic behind the snaps and crashes. When the lines do work, they work well. The quality is good, and they play when you expect them too. The files are also recorded in a similar way as the original EA voicelines (meaning short silence at the beginning and end, similar lenght, etc.). It seems some lines are more prone to crash or snap than others, but we have not been able to find any logic behind this. 

Here is what we have done so far:
- Recorded voicelines in good quality (MP3 or Wav), in a comparable logic to EA's own recordings
- Convert the files to one of the formats BattleForge uses
- Create an  ACK file for the new cards that references the voicelines in the game, using the same template as currently existing units. 
- Try out the voicelines in all kind of settings (at some point we even connected them to the login button to test them in quick succession). 

Current theories we have what are causing these errors:
- Weird sounds and crashes might be caused by wrong size*
- The issues could be caused by the converter/generator of SNR files 
- Corrupting memory allocator should not be possible by any data in any software, so it is a bug in the game.

*With this we dont mean "size" as in file size, but "size'' of memory allocation or "size" of decompressed audio

Which leads us to the current options:
- Use 70% compressed wav. This would mean low quality sounds. We hope the community can improve the wav to snr to have more seamless result. 
- Find someone to reverse engineer this part of BF, and find the issue (Kubik did not yet find the root cause, only that it corrupts memory allocator somehow, which should be almost impossible)
- Wav to snr mp3 branch can be improved / extended by someone (or a coordinated group) who understand coding and/or sound enough to add diagnostic information to the sound itself and find the root cause this way
- Find someone who has dealt with this kind of issue for other EA games, who used the other codecs, and has (or can write) a convertor, and try those. 
- Do not implement sounds at all
 

Please let us know if you have any questions, theories or, preferably, solutions. If there is anything you need for testing, please let us know!

[Majora]

I know this one got overshadowed a bit by the new-patch-hype, but we would greatly appreciate the help from the community for making the implementation of voicelines possible. We already have two cards fully recorded and ready to go, with others in various stages of recordings, but they are currently on hold because of the mentioned issues. 

Please help us out 🙂