Code Base

This section maps the model components onto the code base as reference for development.

Source-Filter Encoder¶

Code	Purpose	Note
Speaker Embedding	Architecture#Disentangled Speech Representation	[[Meta-StyleSpeech]]
F0 Extraction	Architecture#Disentangled Speech Representation	Fundamental Frequency (F0)
VQ-VAE	Architecture#Disentangled Speech Representation	[[VQ-VAE]]
XLS-R	Architecture#Disentangled Speech Representation	Wav2Vec 2.0#XLS-R

Code	Purpose	Note
Diffusion Module	Disentangled Denoising#Training Objective
Score Network	Disentangled Denoising#Reverse Process
Beta	Disentangled Denoising#Training Objective	computes \(\beta_{t}\)
Gamma	Disentangled Denoising#Training Objective	computes \(e^{-\frac{1}{2}\int_{0}^{t}\beta_{s}ds}\)
Forward Diffusion	Disentangled Denoising#Forward Process	computes \(X_{t}\)
Reverse Diffusion	Disentangled Denoising#Reverse Process	computes \(\hat{X_{t}}\)
Diffusion Loss	Architecture#Training and Loss

Code	Purpose	Note
DDDM-VC	Overview
Source-Filter Encoder	Architecture#Source-Filter Encoder
#Source-Filter Decoder	Architecture#Source-Filter Decoder
Total Loss	Architecture#Training and Loss	with [[Prior Mixup]]

Code	Purpose	Note
HIFIGAN	HiFiGAN

Code	Purpose	Note
Data Augmentation	Data augmentation