misaim / ebl-reading Goto Github PK

Currently output file verification is reasonably strict:

re.sub('[^0-9a-zA-Z\.,:%\-_#]+', '', ebl_file['header_3']['filename'])

We should add a new flag and a secondary regex which allows some chars through for OS's where it doesn't matter - i.e Mac OS X and Linux both allow !, ? and ' ' chars in filenames, although windows doesn't.

Could also look at replacing some strings. '/w' is used for shorthand "with", / is used for time sig's and could be "slash" or a dash.

Filename conversion sucks right now. I manually replace every character that I've found which is illegal on my platform (mac) and linux.

Ideally we only replace chars outside of A-Z, a-z, and 0-9.
Some exceptions are

• # (Sharp in music notation)
• \w (Common "with" abbreviation)
• / (Slash? Potentially with a '-'... Used for fake filenames/folders in some old files)
• _ Because why not
• - For the same reason

Should log them maybe.

Testing is... Lackluster. I'd like to have a stack of home-made EBL files with known WAV outputs that we can compare to.

EBL reading sucks presently. mono-support branch kind of improves it, but we still have problems with some headers.

I think they are smaller (or bigger... depends on audio start) than the majority of the dataset. Currently, header files are mostly assumed to be the correct size, while we do check for channel size and read the appropriate amount of bytes - this is improved on mono-support branch.

Probably we should be checking a bunch of headers (automatically) for consistency and spot patterns - like header 3 length in bytes is encoded in header 2, and the data channel sizes are encoded weirdly.

Synth KitSL001.ebl.zip

As per @jackbucinskas suggestion, we should swap to better performance monitoring i.e

start_time = time.perf_counter()
    
for i in range(10):
    print(i)
    
end_time = time.perf_counter()
execution_time = end_time - start_time
print(f"The execution time is: {execution_time}")

In main.py we have these lines (Roughly 199 in latest version):

wav_data = b""
for i in range(0, channel_size, 2):
    wav_data += channel_1_data[i : i + 2]
    wav_data += channel_2_data[i : i + 2]

This combines two continuous sets of bytes into one like a zipper, 2 bytes at a time.

I haven't profiled it, but large files (>1mb) tend to take a while, with this as the only hold up I can see. I know it's slow - it's doing a copy every time!

When running conversions with -p (Preserve Filename) flag, the resulting output has the originals filename (with suffix as .wav).
The metadata-encoded Emulator X-3 filename is lost. Suggestions:

• Append metadata filename to end of original filename
• ...The opposite
• Encode trackname within WAV file

Change handling of input files/folders
https://docs.python.org/3/library/pathlib.html#pathlib.Path.resolve

From @jackbucinskas

i personally don't like so many embedded if and for loops, it makes it hard for someone to pick up the code in the future
i would suggest moving the if / for loop into a separate function with a descriptive function name

He's right, it's a bit of a mess. I was several beers deep. Needs to be re-worked.