added fix for comma instead of dot in floating point numbers

[georgid]

When there is a timestamp stored in a text file with comma instead of dot for dividing the whole number from the mantissa (e.g. 1,23 instead of 1.23), load_delimited throws an error.

[craffel]

Is this common e.g. in Europe? Is mir_eval Americentric? I would otherwise argue that load_delimited isn't meant to do any actual conversion of the values - they should be ready-to-go for int() or float() or whatever. In other words, I would say if this is a bug it's a bug with float(). You could also fix this by passing in an appropriate "converter" for the col in question, e.g. lambda val: float(val.replace(',', '.').

[georgid]

The "converter" fix with the converter is absolutely solving the issue. Otherwise, I dont think that ',' instead of '.' is Europe-related situation. It is very likely to appear when values are exported from excel or other postprocessing editors that use ','-s.