update pre-trained models with audio demos

bobo-paopao · Oct 17, 2021 · f9161cf · f9161cf
1 parent 84fad6f
commit f9161cf
Show file tree

Hide file tree

Showing 107 changed files with 9 additions and 6 deletions.
diff --git a/README.md b/README.md
@@ -6,6 +6,9 @@ PyTorch Implementation of [PortaSpeech: Portable and High-Quality Generative Tex
     <img src="img/model.png" width="80%">
 </p>
 
+## Audio Samples
+Audio samples are available at [/demo](https://github.com/keonlee9420/PortaSpeech/tree/main/demo).
+
 ## Model Size
 | Module | Normal | Small | Normal (paper) | Small (paper) |
 | :----- | :-----: | :-----: | :-----: | :-----: |
@@ -111,11 +114,11 @@ tensorboard --logdir output/log
 ```
 
 to serve TensorBoard on your localhost.
-<!-- The loss curves, synthesized mel-spectrograms, and audios are shown.
+The loss curves, synthesized mel-spectrograms, and audios are shown.
 
 ![](./img/tensorboard_loss.png)
 ![](./img/tensorboard_spec.png)
-![](./img/tensorboard_audio.png) -->
+![](./img/tensorboard_audio.png)
 
 # Notes
 

diff --git a/config/LJSpeech/model_small.yaml b/config/LJSpeech/model_small.yaml
@@ -11,7 +11,7 @@ variational_generator:
   encoder_layer: 8
   decoder_layer: 3
   conv_kernel_size: 3
-  conv_stride_size: 4
+  conv_stride_size: 2
   encoder_decoder_hidden: 128
   # encoder_decoder_dropout: 0.2
   latent_hidden: 16

diff --git a/config/LJSpeech/preprocess.yaml b/config/LJSpeech/preprocess.yaml
@@ -12,8 +12,8 @@ preprocessing:
   text:
     text_cleaners: ["english_cleaners"]
     language: "en"
-    sub_divide_word: True
-    max_phoneme_num: 7
+    sub_divide_word: True # subdivide long words into smaller pieces
+    max_phoneme_num: 7 # threshold for subdivision
   audio:
     sampling_rate: 22050
     max_wav_value: 32768.0

diff --git a/config/LJSpeech/train.yaml b/config/LJSpeech/train.yaml
@@ -15,7 +15,7 @@ optimizer:
   grad_clip_thresh: 1.0
   grad_acc_step: 1
   warm_up_step: 4000
-  anneal_steps: [100000, 200000, 300000]
+  anneal_steps: [175000, 250000, 300000]
   anneal_rate: 0.3
 step:
   total_step: 500000

diff --git a/demo/LJSpeech/normal/100k/LJ001-0165.png b/demo/LJSpeech/normal/100k/LJ001-0165.png
diff --git a/demo/LJSpeech/normal/100k/LJ001-0165.wav b/demo/LJSpeech/normal/100k/LJ001-0165.wav
diff --git a/demo/LJSpeech/normal/100k/LJ001-0174.png b/demo/LJSpeech/normal/100k/LJ001-0174.png
diff --git a/demo/LJSpeech/normal/100k/LJ001-0174.wav b/demo/LJSpeech/normal/100k/LJ001-0174.wav
diff --git a/demo/LJSpeech/normal/100k/LJ002-0002.png b/demo/LJSpeech/normal/100k/LJ002-0002.png
diff --git a/demo/LJSpeech/normal/100k/LJ002-0002.wav b/demo/LJSpeech/normal/100k/LJ002-0002.wav
diff --git a/demo/LJSpeech/normal/100k/LJ011-0269.png b/demo/LJSpeech/normal/100k/LJ011-0269.png
diff --git a/demo/LJSpeech/normal/100k/LJ011-0269.wav b/demo/LJSpeech/normal/100k/LJ011-0269.wav
diff --git a/demo/LJSpeech/normal/100k/LJ017-0025.png b/demo/LJSpeech/normal/100k/LJ017-0025.png
diff --git a/demo/LJSpeech/normal/100k/LJ017-0025.wav b/demo/LJSpeech/normal/100k/LJ017-0025.wav
diff --git a/demo/LJSpeech/normal/100k/LJ028-0403.png b/demo/LJSpeech/normal/100k/LJ028-0403.png
diff --git a/demo/LJSpeech/normal/100k/LJ028-0403.wav b/demo/LJSpeech/normal/100k/LJ028-0403.wav
diff --git a/demo/LJSpeech/normal/100k/LJ045-0141.png b/demo/LJSpeech/normal/100k/LJ045-0141.png
diff --git a/demo/LJSpeech/normal/100k/LJ045-0141.wav b/demo/LJSpeech/normal/100k/LJ045-0141.wav
diff --git a/...e remembered is that the ornament, whatever it is, whether picture or patte.png b/...e remembered is that the ornament, whatever it is, whether picture or patte.png
diff --git a/...point to be remembered is that the ornament, whatever it is, whether picture or patte.wav b/...point to be remembered is that the ornament, whatever it is, whether picture or patte.wav
diff --git a/... are of an earlier character, though they also much resemble flemish or col.png b/... are of an earlier character, though they also much resemble flemish or col.png
diff --git a/... own types are of an earlier character, though they also much resemble flemish or col.wav b/... own types are of an earlier character, though they also much resemble flemish or col.wav
diff --git a/...gements are to be found in mr. neilds state of prisons in england, scotland.png b/...gements are to be found in mr. neilds state of prisons in england, scotland.png
diff --git a/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav b/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav
diff --git a/demo/LJSpeech/normal/125k/LJ001-0165.png b/demo/LJSpeech/normal/125k/LJ001-0165.png
diff --git a/demo/LJSpeech/normal/125k/LJ001-0165.wav b/demo/LJSpeech/normal/125k/LJ001-0165.wav
diff --git a/demo/LJSpeech/normal/125k/LJ001-0174.png b/demo/LJSpeech/normal/125k/LJ001-0174.png
diff --git a/demo/LJSpeech/normal/125k/LJ001-0174.wav b/demo/LJSpeech/normal/125k/LJ001-0174.wav
diff --git a/demo/LJSpeech/normal/125k/LJ002-0002.png b/demo/LJSpeech/normal/125k/LJ002-0002.png
diff --git a/demo/LJSpeech/normal/125k/LJ002-0002.wav b/demo/LJSpeech/normal/125k/LJ002-0002.wav
diff --git a/demo/LJSpeech/normal/125k/LJ011-0269.png b/demo/LJSpeech/normal/125k/LJ011-0269.png
diff --git a/demo/LJSpeech/normal/125k/LJ011-0269.wav b/demo/LJSpeech/normal/125k/LJ011-0269.wav
diff --git a/demo/LJSpeech/normal/125k/LJ017-0025.png b/demo/LJSpeech/normal/125k/LJ017-0025.png
diff --git a/demo/LJSpeech/normal/125k/LJ017-0025.wav b/demo/LJSpeech/normal/125k/LJ017-0025.wav
diff --git a/demo/LJSpeech/normal/125k/LJ028-0403.png b/demo/LJSpeech/normal/125k/LJ028-0403.png
diff --git a/demo/LJSpeech/normal/125k/LJ028-0403.wav b/demo/LJSpeech/normal/125k/LJ028-0403.wav
diff --git a/demo/LJSpeech/normal/125k/LJ045-0141.png b/demo/LJSpeech/normal/125k/LJ045-0141.png
diff --git a/demo/LJSpeech/normal/125k/LJ045-0141.wav b/demo/LJSpeech/normal/125k/LJ045-0141.wav
diff --git a/... remembered is that the ornament, whatever it is, whether picture or patter.png b/... remembered is that the ornament, whatever it is, whether picture or patter.png
diff --git a/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav b/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav
diff --git a/... are of an earlier character, though they also much resemble flemish or col.png b/... are of an earlier character, though they also much resemble flemish or col.png
diff --git a/... own types are of an earlier character, though they also much resemble flemish or col.wav b/... own types are of an earlier character, though they also much resemble flemish or col.wav
diff --git a/...gements are to be found in mr. neilds state of prisons in england, scotland.png b/...gements are to be found in mr. neilds state of prisons in england, scotland.png
diff --git a/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav b/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav
diff --git a/demo/LJSpeech/small/150k/LJ001-0165.png b/demo/LJSpeech/small/150k/LJ001-0165.png
diff --git a/demo/LJSpeech/small/150k/LJ001-0165.wav b/demo/LJSpeech/small/150k/LJ001-0165.wav
diff --git a/demo/LJSpeech/small/150k/LJ001-0174.png b/demo/LJSpeech/small/150k/LJ001-0174.png
diff --git a/demo/LJSpeech/small/150k/LJ001-0174.wav b/demo/LJSpeech/small/150k/LJ001-0174.wav
diff --git a/demo/LJSpeech/small/150k/LJ002-0002.png b/demo/LJSpeech/small/150k/LJ002-0002.png
diff --git a/demo/LJSpeech/small/150k/LJ002-0002.wav b/demo/LJSpeech/small/150k/LJ002-0002.wav
diff --git a/demo/LJSpeech/small/150k/LJ011-0269.png b/demo/LJSpeech/small/150k/LJ011-0269.png
diff --git a/demo/LJSpeech/small/150k/LJ011-0269.wav b/demo/LJSpeech/small/150k/LJ011-0269.wav
diff --git a/demo/LJSpeech/small/150k/LJ017-0025.png b/demo/LJSpeech/small/150k/LJ017-0025.png
diff --git a/demo/LJSpeech/small/150k/LJ017-0025.wav b/demo/LJSpeech/small/150k/LJ017-0025.wav
diff --git a/demo/LJSpeech/small/150k/LJ028-0403.png b/demo/LJSpeech/small/150k/LJ028-0403.png
diff --git a/demo/LJSpeech/small/150k/LJ028-0403.wav b/demo/LJSpeech/small/150k/LJ028-0403.wav
diff --git a/demo/LJSpeech/small/150k/LJ045-0141.png b/demo/LJSpeech/small/150k/LJ045-0141.png
diff --git a/demo/LJSpeech/small/150k/LJ045-0141.wav b/demo/LJSpeech/small/150k/LJ045-0141.wav
diff --git a/... remembered is that the ornament, whatever it is, whether picture or patter.png b/... remembered is that the ornament, whatever it is, whether picture or patter.png
diff --git a/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav b/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav
diff --git a/... are of an earlier character, though they also much resemble flemish or col.png b/... are of an earlier character, though they also much resemble flemish or col.png
diff --git a/... own types are of an earlier character, though they also much resemble flemish or col.wav b/... own types are of an earlier character, though they also much resemble flemish or col.wav
diff --git a/...gements are to be found in mr. neilds state of prisons in england, scotland.png b/...gements are to be found in mr. neilds state of prisons in england, scotland.png
diff --git a/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav b/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav
diff --git a/demo/LJSpeech/small/175k/LJ001-0165.png b/demo/LJSpeech/small/175k/LJ001-0165.png
diff --git a/demo/LJSpeech/small/175k/LJ001-0165.wav b/demo/LJSpeech/small/175k/LJ001-0165.wav
diff --git a/demo/LJSpeech/small/175k/LJ001-0174.png b/demo/LJSpeech/small/175k/LJ001-0174.png
diff --git a/demo/LJSpeech/small/175k/LJ001-0174.wav b/demo/LJSpeech/small/175k/LJ001-0174.wav
diff --git a/demo/LJSpeech/small/175k/LJ002-0002.png b/demo/LJSpeech/small/175k/LJ002-0002.png
diff --git a/demo/LJSpeech/small/175k/LJ002-0002.wav b/demo/LJSpeech/small/175k/LJ002-0002.wav
diff --git a/demo/LJSpeech/small/175k/LJ011-0269.png b/demo/LJSpeech/small/175k/LJ011-0269.png
diff --git a/demo/LJSpeech/small/175k/LJ011-0269.wav b/demo/LJSpeech/small/175k/LJ011-0269.wav
diff --git a/demo/LJSpeech/small/175k/LJ017-0025.png b/demo/LJSpeech/small/175k/LJ017-0025.png
diff --git a/demo/LJSpeech/small/175k/LJ017-0025.wav b/demo/LJSpeech/small/175k/LJ017-0025.wav
diff --git a/demo/LJSpeech/small/175k/LJ028-0403.png b/demo/LJSpeech/small/175k/LJ028-0403.png
diff --git a/demo/LJSpeech/small/175k/LJ028-0403.wav b/demo/LJSpeech/small/175k/LJ028-0403.wav
diff --git a/demo/LJSpeech/small/175k/LJ045-0141.png b/demo/LJSpeech/small/175k/LJ045-0141.png
diff --git a/demo/LJSpeech/small/175k/LJ045-0141.wav b/demo/LJSpeech/small/175k/LJ045-0141.wav
diff --git a/... remembered is that the ornament, whatever it is, whether picture or patter.png b/... remembered is that the ornament, whatever it is, whether picture or patter.png
diff --git a/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav b/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav
diff --git a/... are of an earlier character, though they also much resemble flemish or col.png b/... are of an earlier character, though they also much resemble flemish or col.png
diff --git a/... own types are of an earlier character, though they also much resemble flemish or col.wav b/... own types are of an earlier character, though they also much resemble flemish or col.wav
diff --git a/...gements are to be found in mr. neilds state of prisons in england, scotland.png b/...gements are to be found in mr. neilds state of prisons in england, scotland.png
diff --git a/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav b/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav
diff --git a/demo/LJSpeech/small/200k/LJ001-0165.png b/demo/LJSpeech/small/200k/LJ001-0165.png
diff --git a/demo/LJSpeech/small/200k/LJ001-0165.wav b/demo/LJSpeech/small/200k/LJ001-0165.wav
diff --git a/demo/LJSpeech/small/200k/LJ001-0174.png b/demo/LJSpeech/small/200k/LJ001-0174.png
diff --git a/demo/LJSpeech/small/200k/LJ001-0174.wav b/demo/LJSpeech/small/200k/LJ001-0174.wav
diff --git a/demo/LJSpeech/small/200k/LJ002-0002.png b/demo/LJSpeech/small/200k/LJ002-0002.png
diff --git a/demo/LJSpeech/small/200k/LJ002-0002.wav b/demo/LJSpeech/small/200k/LJ002-0002.wav
diff --git a/demo/LJSpeech/small/200k/LJ011-0269.png b/demo/LJSpeech/small/200k/LJ011-0269.png
diff --git a/demo/LJSpeech/small/200k/LJ011-0269.wav b/demo/LJSpeech/small/200k/LJ011-0269.wav
diff --git a/demo/LJSpeech/small/200k/LJ017-0025.png b/demo/LJSpeech/small/200k/LJ017-0025.png
diff --git a/demo/LJSpeech/small/200k/LJ017-0025.wav b/demo/LJSpeech/small/200k/LJ017-0025.wav
diff --git a/demo/LJSpeech/small/200k/LJ028-0403.png b/demo/LJSpeech/small/200k/LJ028-0403.png
diff --git a/demo/LJSpeech/small/200k/LJ028-0403.wav b/demo/LJSpeech/small/200k/LJ028-0403.wav
diff --git a/demo/LJSpeech/small/200k/LJ045-0141.png b/demo/LJSpeech/small/200k/LJ045-0141.png
diff --git a/demo/LJSpeech/small/200k/LJ045-0141.wav b/demo/LJSpeech/small/200k/LJ045-0141.wav
diff --git a/... remembered is that the ornament, whatever it is, whether picture or patter.png b/... remembered is that the ornament, whatever it is, whether picture or patter.png
diff --git a/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav b/...oint to be remembered is that the ornament, whatever it is, whether picture or patter.wav
diff --git a/... are of an earlier character, though they also much resemble flemish or col.png b/... are of an earlier character, though they also much resemble flemish or col.png
diff --git a/... own types are of an earlier character, though they also much resemble flemish or col.wav b/... own types are of an earlier character, though they also much resemble flemish or col.wav
diff --git a/...gements are to be found in mr. neilds state of prisons in england, scotland.png b/...gements are to be found in mr. neilds state of prisons in england, scotland.png
diff --git a/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav b/... the arrangements are to be found in mr. neilds state of prisons in england, scotland.wav
diff --git a/img/tensorboard_audio.png b/img/tensorboard_audio.png
diff --git a/img/tensorboard_loss.png b/img/tensorboard_loss.png
diff --git a/img/tensorboard_spec.png b/img/tensorboard_spec.png