A generic evaluation-time tokenizer, which leverages built-in tokenizers in sacreBLEU (https://github.com/mjpost/sacrebleu). It additionally provi

:class:`transformers.LogitsProcessor` that enforces that when specified sequences appear, stop geration. Args: stop_words_ids (:obj:

Score specifically for pitch detection -- converts all pitches to chroma first. This score ignores octave errors in pitch classification.

AUCROC (macro mode) is calculated per class followed by averaging across the classes

DPrime is calculated per class followed by averaging across the classes Code adapted from code provided by Eduoard Fonseca.

event-based scores - the ground truth and system output are compared at event instance level; See https://tut-arg.github.io/sed_eval/gen

Average Precision is calculated in macro mode which calculates AP at a class level followed by macro-averaging across the classes.

QWen tokenizer.

A simple abstract base class for score functions

segment-based scores - the ground truth and system output are compared in a fixed time grid; sound events are marked as active or inactive in

Scores for sound event detection tasks using sed_eval