create project

caizexin · Oct 4, 2020 · 0910f62 · 0910f62
1 parent f40ee23
commit 0910f62
Show file tree

Hide file tree

Showing 223 changed files with 172,499 additions and 0 deletions.
diff --git a/.gitignore b/.gitignore
@@ -0,0 +1,14 @@
+synthesizer/saved_models
+synthesizer/pre_trained
+feedback_synthesizer/saved_models
+vocoder/saved_models
+vocoder/pretrained
+deep_speaker/data_vox12/vox12_train_aug.csv
+deep_speaker/vox12_resnet34
+deep_speaker/log
+deep_speaker/database
+datasets/vctk/wavs
+datasets/vctk/synthesizer/audio
+datasets/vctk/synthesizer/embeds
+datasets/vctk/synthesizer/mels
+!.gitignore
diff --git a/README b/README
@@ -0,0 +1,17 @@
+1. speaker verification model
+    - cd deep_speaker
+    - CUDA_VISIBLE_DEVICES=1 python train.py
+
+1. synthesizer without feedback control:
+    - extract audio feature using process_audio.ipynb (remember to split validation set and test set out for performance evaluation)
+    - extract embedding using script deep_speaker/get_gvector.ipynb
+    - train baseline multispeaker TTS system using synthesizer_train.py (CUDA_VISIBLE_DEVICES=1 python synthesizer_train.py vctk datasets/vctk/synthesizer)
+    - validate and synthesize result using syn.ipynb at any time during training
+    - using pre-trained WaveRNN vocoder in vocoder/pretrained, or train new one using vocoder_train.py. (CUDA_VISIBLE_DEVICES=1 python vocoder_train.py -g --syn_dir datasets/vctk/synthesizer  vctk datasets/vctk) For better performance, please use GTA Mel-spectrogram obtained by vocoder_preprocess.py. 
+
+2. synthesizer with feedback constraint: 
+    - load two model (speaker verification model and multispeaker synthesizer) by changing the path in feedback_synthesizer/hparams.py
+    - CUDA_VISIBLE_DEVICES=1 python fc_synthesizer_train.py
+    - evaluate with feedback_syn.ipynb
+
+
diff --git a/datasets/vctk/metadata.csv b/datasets/vctk/metadata.csv
diff --git a/datasets/vctk/synthesizer/all.txt b/datasets/vctk/synthesizer/all.txt
diff --git a/datasets/vctk/synthesizer/test.txt b/datasets/vctk/synthesizer/test.txt
diff --git a/datasets/vctk/synthesizer/train.txt b/datasets/vctk/synthesizer/train.txt
diff --git a/datasets/vctk/synthesizer/val.txt b/datasets/vctk/synthesizer/val.txt
diff --git a/deep_speaker/.ipynb_checkpoints/eer_cal-checkpoint.ipynb b/deep_speaker/.ipynb_checkpoints/eer_cal-checkpoint.ipynb
@@ -0,0 +1,6 @@
+{
+ "cells": [],
+ "metadata": {},
+ "nbformat": 4,
+ "nbformat_minor": 2
+}