| Thu, 19 Sep 2019 15:17:56 +0100 |
Chris Cannam |
Introduce time-frequency box model
time-frequency-boxes
|
| Tue, 17 Sep 2019 11:21:33 +0100 |
Chris Cannam |
More graceful handling of failure to construct FFT models in the case where the source model has already been deleted before this occurs
|
| Tue, 17 Sep 2019 10:42:55 +0100 |
Chris Cannam |
Fix calculation of binary name (don't include full path!)
|
| Tue, 17 Sep 2019 09:28:27 +0100 |
Chris Cannam |
Lib dir name is usually the same as the binary name, not the formal application name
|
| Mon, 16 Sep 2019 14:23:17 +0100 |
Chris Cannam |
Update helper dir path to look in equivalent of /usr/lib/xxxx on Linux
|
| Thu, 12 Sep 2019 11:52:19 +0100 |
Chris Cannam |
Support optional max frequency setting, useful when we want to store caches of very constrained frequency ranges (as in melodic-range spectrogram, potentially)
|
| Wed, 11 Sep 2019 13:20:40 +0100 |
Chris Cannam |
Update macOS memory size check - the original sysctl was returning far too small a value for modern machines (because bounded to a 32-bit int)
|
| Wed, 11 Sep 2019 11:19:27 +0100 |
Chris Cannam |
Replace the model used for the cache part of the peak-cache model with a simple vector of vectors. Avoids unnecessary locking in a class that is not thread-safe in any case. Also record whether the final column is actually truncated, rather than risk possible backward seeks to re-read it in the case where it simply might be
|
| Tue, 10 Sep 2019 16:34:47 +0100 |
Chris Cannam |
Split EditableDenseThreeDimensionalModel into explicitly compressed and uncompressed variants. Simplifies the uncompressed version, and we may want to consider whether we need the compressed one at all.
|
| Mon, 09 Sep 2019 10:25:16 +0100 |
Chris Cannam |
Don't even ask to unload a library if we know about other plugins that are loaded from it. Plus debug tweaks
|
| Wed, 04 Sep 2019 17:31:02 +0100 |
Chris Cannam |
Some debug tweaks
|
| Thu, 15 Aug 2019 18:18:03 +0100 |
Chris Cannam |
Optionally record relative pitch
|
| Wed, 14 Aug 2019 13:54:23 +0100 |
Chris Cannam |
This output is too annoying, in the perfectly innocuous case of reading from an aggregate model whose components are different lengths
|
| Wed, 14 Aug 2019 11:50:56 +0100 |
Chris Cannam |
Be stricter - require org name as well as app name before allowing ResourceFinder to be used. This is because the application name may have a default, so we could end up with inconsistent results from using before & after these things have been set
|
