romanUmpire.py's analysis as indicator for alignment?

[AndyWeasley2004]

Hi,

Really, thanks for collecting and sharing such a large corpus of roman numeral annotations and the source of score files. If I'm looking for music score and score-aligned analysis, are the results (pitch match) produced by romanUmpire reliable? What is an estimated threshold that can imply the analysis is not aligned at all? I find many pieces have pitch match < 60%. Is it normal, based on your knowledge, to this database (as I modified the Code scripts locally for my data organization and format)?

Thank you for your help!

[MarkGotham]

Hi @AndyWeasley2004

Thanks for this.

@napulen and I had a big clean up at some point (reported in issues and commits here, I think) and I thought we got everything over 80% at that time.

From your message, it sounds like not, or perhaps that more problematic cases have since been added. I'd draw a distinction here between local and converted corpora. See discussion in a previous issue (#98, I think). It may be that the clean up to over 80% took place before I included some of the DCML conversions.

In any case, I agree that the lower match indicates error. I'm happy to take a look at any examples and/or PRs, e.g. for individual, local cases if you find any typos and/or improvements to the DCML->m21 conversion routines as needed. Also happy to continue over email if you prefer (I'm @ KCL).

Thanks again