assign_word_speakers fix

[grazder]

Hello! Seems like here is mistake. We should look for closest instead of sum.

Simple counterexample:

predicted_sample - [0,1]

diarization_segments:
SPEAKER_00: [2, 3], [4, 5]
SPEAKER_01: [3, 4]

For SPEAKER_00 value of dia_tmp.groupby("speaker")["intersection"].sum().sort_values(ascending=False).index[0] will be sum([-1, -3]) = -4. For SPEAKER_01 value will be sum([-2]) = -2. (unless I made a mistake in my calculations in my head)

And therefore we will choose SPEAKER_01, even if it is not the closest

[Barabazs]

@m-bain could you have a look at this please?