Voice annotation quality consistency detection method

Abstract

The invention provides a voice annotation quality consistency detection method. The voice annotation quality consistency detection method comprises the following steps of (1) annotating a file to be in a TextGrid format, dividing annotation forms into three types and designing three corresponding consistency detection formulas according to the three different annotation forms; (2) detecting consistency of a file comprising six layers of annotation information, wherein the six layers of information respectively is a polysyllable layer (PY layer), an initial consonant and simple vowel layer (SY layer), an unvoiced sound and voice sound mute layer (SUV layer), an auxiliary language information layer (PARAL layer), an emotion layer (EMO layer), a stress index layer (ST layer); (3) selecting the corresponding consistency detection formulas according to the annotation forms of the annotation layer; (4) setting the time error T0, enabling annotation results to be consistent if annotation contents of two annotator are identical and the time error or less than and equal to T0, enabling the annotation results to be inconsistent and sequentially calculating the consistency of the layers. The voice annotation quality consistency detection method provides basis for speech database annotation quality detection and is effectively for detection of automatic voice annotation quality and manual annotation quality.

Description

technical field [0001] The invention relates to a consistency detection method for checking the quality of voice annotations. The method can compare two marked files to check the consistency of the marking results, and belongs to the field of voice signal processing. Background technique [0002] Speech annotation means that the annotator divides and annotates the speech according to certain annotation rules, including syllables, initials, finals, unvoiced sounds, voiced sounds, mutes, paralinguistic information, stress information, etc. Consistency detection means that different voice annotators annotate the same voice, then compare the annotated results, observe and analyze the sameness and difference of the annotated results. The consistency of manual voice annotation is an important index to evaluate the voice quality and the completeness of the annotation system. In order to ensure the labeling quality of the emotional voice database and to check the integrity of the l...

AI Technical Summary

Problems solved by technology

At present, voice annotation at home and abroad is carried out according to their own research needs, and the consistency detection method is even more lacking. There is no unified annotation standard and detection method.

Images