Abstract
PROBLEM TO BE SOLVED: To provide an evaluation method, an evaluation device, and an evaluation program that can accurately associate operation modes with each cluster for feature amounts extracted from voice data.
SOLUTION: An evaluation method includes processing of: extracting feature amounts at each clock time from voice data; generating a plurality of clusters for the feature amounts by clustering the extracted feature amounts; calculating the transitional probability from each of the plurality of clusters to other clusters, on the basis of the deviation of the clusters to which the feature amounts at each clock time belong; and associating each of the plurality of clusters with one of either first data indicating a subject is under steady operation, second data indicating the subject is under unsteady operation, or third data indicating the subject is under suspension, on the basis of the transitional probability from each of the plurality of clusters to other clusters.
SOLUTION: An evaluation method includes processing of: extracting feature amounts at each clock time from voice data; generating a plurality of clusters for the feature amounts by clustering the extracted feature amounts; calculating the transitional probability from each of the plurality of clusters to other clusters, on the basis of the deviation of the clusters to which the feature amounts at each clock time belong; and associating each of the plurality of clusters with one of either first data indicating a subject is under steady operation, second data indicating the subject is under unsteady operation, or third data indicating the subject is under suspension, on the basis of the transitional probability from each of the plurality of clusters to other clusters.
Translated title of the contribution | Evaluation method, evaluation device, and evaluation program: JP 2018185711 A |
---|---|
Original language | Japanese |
Patent number | JP 2018185711 A |
IPC | G06F 16/65,G01H 17/00 |
Filing date | 27/04/17 |
Publication status | Published - 22 Nov 2018 |
Externally published | Yes |