Evaluation methods in SpeechSynthesis(including VC)

subjective

CMOS : https://ecs.utdallas.edu/loizou/cimplants/quality_assessment_chapter.pdf

Objective evaluation in speech synthesis

display overall

👉https://arxiv.org/pdf/2104.00355.pdf

defination in papers

MSD, F0 RMSE, F0 corr, GPE, FPE

MCD, GPE, VDE, FFE MCD

training

MCD

definination

http://www1.se.cuhk.edu.hk/~hccl/publications/pub/2016_paper_297.pdf section 4.2

code

mcd(34 mcep) other implement

normal value range

log dB 4.4

F0 RMSE/ corr/ VDE/ FFE

GPE

normal value range

RMSE 22.386

VUV

test

common

content

PER, WER

speaker similarity

https://github.com/resemble-ai/Resemblyzer tsne

in paper Building Bilingual and Code-Switched Voice Conversion with Limited Training Data Using Embedding Consistency Loss