Dan Ellis

OK, I just added a HAVE_FFMPEG flag to the top of audio_read.py. If you set this to False, the FFMPEG code is skipped and instead sound files are read with...

Any data on how much this helps? It seems like this is going to be single-digit percent in hash utilization, and it that doesn’t directly translate into improved recognition or...

More fully occupying the hash space is nice, but it seems second-order. It certainly doesn't translate into needing fewer hashes to maintain match accuracy. I don't mean to dismiss your...

I had difficulty getting the code to run under Python3; specifically, I couldn't get librosa to install under python 3.7 MacOS mojave. But when I replaced librosa.stft with a equivalent...

I just pushed a bunch of changes to fix Python 3.7 compatibility; I'm hoping this will fix your issues.

Sounds good. I actually attempted to remove the dependency on librosa by adding stft.py; it's still referenced in the display stuff, but I assume most people don't use that. DAn....

I actually didn't know about audfprint-gui until this thread. Supporting more options is more of a question for mitin001. Best, DAn.

The matching is intrinsically 1 query against the whole reference set, but each match is quite quick so you can run through a lot of queries in a loop. The...

There's a field, Matcher.max_alignments_per_id, which controls how many matches are returned for each ref item. It's 100 by default and there's no way to control it at present. You could...

Glad it worked! max_alignments_per_id = 0 seems like an odd choice, I was thinking max_alignments_per_id = 1 (we want at most 1 hit per reference item), but as long as...