Merge pull request #822 from babysor/restruct-project

Restruct project

.gitignore

@@ -14,9 +14,9 @@
 *.bcf
 *.toc
 *.sh
-*/saved_models
-!vocoder/saved_models/pretrained/**
-!encoder/saved_models/pretrained.pt
+data/ckpt
+!data/ckpt/vocoder/pretrained/**
+!data/ckpt/encoder/pretrained.pt
 wavs
 log
 !/docker-entrypoint.sh

.vscode/launch.json

@@ -15,7 +15,8 @@
       "name": "Python: Vocoder Preprocess",
       "type": "python",
       "request": "launch",
-      "program": "vocoder_preprocess.py",
+      "program": "control\\cli\\vocoder_preprocess.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["..\\audiodata"]
     },
@@ -23,7 +24,8 @@
       "name": "Python: Vocoder Train",
       "type": "python",
       "request": "launch",
-      "program": "vocoder_train.py",
+      "program": "control\\cli\\vocoder_train.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["dev", "..\\audiodata"]
     },
@@ -32,6 +34,7 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata"]
     },
@@ -40,6 +43,7 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata","-vc"]
     },
@@ -47,9 +51,9 @@
       "name": "Python: Synth Train",
       "type": "python",
       "request": "launch",
-      "program": "synthesizer_train.py",
+      "program": "train.py",
       "console": "integratedTerminal",
-      "args": ["my_run", "..\\"]
+      "args": ["--type", "synth", "..\\audiodata\\SV2TTS\\synthesizer"]
     },
     {
       "name": "Python: PPG Convert",

archived_untest_files/demo_cli.py

@@ -1,9 +1,9 @@
-from encoder.params_model import model_embedding_size as speaker_embedding_size
+from models.encoder.params_model import model_embedding_size as speaker_embedding_size
 from utils.argutils import print_args
 from utils.modelutils import check_model_paths
-from synthesizer.inference import Synthesizer
-from encoder import inference as encoder
-from vocoder import inference as vocoder
+from models.synthesizer.inference import Synthesizer
+from models.encoder import inference as encoder
+from models.vocoder import inference as vocoder
 from pathlib import Path
 import numpy as np
 import soundfile as sf

control/cli/encoder_preprocess.py

@@ -1,7 +1,10 @@
-from encoder.preprocess import preprocess_librispeech, preprocess_voxceleb1, preprocess_voxceleb2, preprocess_aidatatang_200zh
-from utils.argutils import print_args
-from pathlib import Path
 import argparse
+from pathlib import Path
+
+from models.encoder.preprocess import (preprocess_aidatatang_200zh,
+                                preprocess_librispeech, preprocess_voxceleb1,
+                                preprocess_voxceleb2)
+from utils.argutils import print_args
 
 if __name__ == "__main__":
     class MyFormatter(argparse.ArgumentDefaultsHelpFormatter, argparse.RawDescriptionHelpFormatter):

control/cli/encoder_train.py

@@ -1,5 +1,5 @@
 from utils.argutils import print_args
-from encoder.train import train
+from models.encoder.train import train
 from pathlib import Path
 import argparse
 

control/cli/ppg2mel_train.py

@@ -2,8 +2,8 @@ import sys
 import torch
 import argparse
 import numpy as np
-from utils.load_yaml import HpsYaml
-from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+from utils.hparams import HpsYaml
+from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
 
 # For reproducibility, comment these may speed up training
 torch.backends.cudnn.deterministic = True

control/cli/pre4ppg.py

@@ -1,7 +1,7 @@
 from pathlib import Path
 import argparse
 
-from ppg2mel.preprocess import preprocess_dataset
+from models.ppg2mel.preprocess import preprocess_dataset
 from pathlib import Path
 import argparse
 

control/cli/synthesizer_train.py

@@ -1,10 +1,9 @@
-from synthesizer.hparams import hparams
-from synthesizer.train import train
+from models.synthesizer.hparams import hparams
+from models.synthesizer.train import train
 from utils.argutils import print_args
 import argparse
 
-
-if __name__ == "__main__":
+def new_train():
     parser = argparse.ArgumentParser()
     parser.add_argument("run_id", type=str, help= \
         "Name for this model instance. If a model state from the same run ID was previously "
@@ -13,7 +12,7 @@ if __name__ == "__main__":
     parser.add_argument("syn_dir", type=str, default=argparse.SUPPRESS, help= \
         "Path to the synthesizer directory that contains the ground truth mel spectrograms, "
         "the wavs and the embeds.")
-    parser.add_argument("-m", "--models_dir", type=str, default="synthesizer/saved_models/", help=\
+    parser.add_argument("-m", "--models_dir", type=str, default=f"data/ckpt/synthesizer/", help=\
         "Path to the output directory that will contain the saved model weights and the logs.")
     parser.add_argument("-s", "--save_every", type=int, default=1000, help= \
         "Number of steps between updates of the model on the disk. Set to 0 to never save the "
@@ -28,10 +27,14 @@ if __name__ == "__main__":
     parser.add_argument("--hparams", default="",
                         help="Hyperparameter overrides as a comma-separated list of name=value "
 							 "pairs")
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
     print_args(args, parser)
 
     args.hparams = hparams.parse(args.hparams)
 
     # Run the training
     train(**vars(args))
+
+
+if __name__ == "__main__":
+    new_train()

control/cli/train_ppg2mel.py

@@ -0,0 +1,66 @@
+import sys
+import torch
+import argparse
+import numpy as np
+from utils.hparams import HpsYaml
+from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+
+# For reproducibility, comment these may speed up training
+torch.backends.cudnn.deterministic = True
+torch.backends.cudnn.benchmark = False
+
+def main():
+    # Arguments
+    parser = argparse.ArgumentParser(description=
+            'Training PPG2Mel VC model.')
+    parser.add_argument('--config', type=str, 
+                        help='Path to experiment config, e.g., config/vc.yaml')
+    parser.add_argument('--name', default=None, type=str, help='Name for logging.')
+    parser.add_argument('--logdir', default='log/', type=str,
+                        help='Logging path.', required=False)
+    parser.add_argument('--ckpdir', default='ppg2mel/saved_models/', type=str,
+                        help='Checkpoint path.', required=False)
+    parser.add_argument('--outdir', default='result/', type=str,
+                        help='Decode output path.', required=False)
+    parser.add_argument('--load', default=None, type=str,
+                        help='Load pre-trained model (for training only)', required=False)
+    parser.add_argument('--warm_start', action='store_true',
+                        help='Load model weights only, ignore specified layers.')
+    parser.add_argument('--seed', default=0, type=int,
+                        help='Random seed for reproducable results.', required=False)
+    parser.add_argument('--njobs', default=8, type=int,
+                        help='Number of threads for dataloader/decoding.', required=False)
+    parser.add_argument('--cpu', action='store_true', help='Disable GPU training.')
+    parser.add_argument('--no-pin', action='store_true',
+                        help='Disable pin-memory for dataloader')
+    parser.add_argument('--test', action='store_true', help='Test the model.')
+    parser.add_argument('--no-msg', action='store_true', help='Hide all messages.')
+    parser.add_argument('--finetune', action='store_true', help='Finetune model')
+    parser.add_argument('--oneshotvc', action='store_true', help='Oneshot VC model')
+    parser.add_argument('--bilstm', action='store_true', help='BiLSTM VC model')
+    parser.add_argument('--lsa', action='store_true', help='Use location-sensitive attention (LSA)')
+
+    ###
+    paras = parser.parse_args()
+    setattr(paras, 'gpu', not paras.cpu)
+    setattr(paras, 'pin_memory', not paras.no_pin)
+    setattr(paras, 'verbose', not paras.no_msg)
+    # Make the config dict dot visitable
+    config = HpsYaml(paras.config)
+
+    np.random.seed(paras.seed)
+    torch.manual_seed(paras.seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(paras.seed)
+
+    print(">>> OneShot VC training ...")
+    mode = "train"
+    solver = Solver(config, paras, mode)
+    solver.load_data()
+    solver.set_model()
+    solver.exec()
+    print(">>> Oneshot VC train finished!")
+    sys.exit(0)
+
+if __name__ == "__main__":
+    main()   

control/cli/vocoder_preprocess.py

@@ -1,5 +1,5 @@
-from synthesizer.synthesize import run_synthesis
-from synthesizer.hparams import hparams
+from models.synthesizer.synthesize import run_synthesis
+from models.synthesizer.hparams import hparams
 from utils.argutils import print_args
 import argparse
 import os

control/cli/vocoder_train.py

@@ -1,7 +1,7 @@
 from utils.argutils import print_args
-from vocoder.wavernn.train import train
-from vocoder.hifigan.train import train as train_hifigan
-from vocoder.fregan.train import train as train_fregan
+from models.vocoder.wavernn.train import train
+from models.vocoder.hifigan.train import train as train_hifigan
+from models.vocoder.fregan.train import train as train_fregan
 from utils.util import AttrDict
 from pathlib import Path
 import argparse

control/mkgui/app.py

@@ -2,22 +2,22 @@ from pydantic import BaseModel, Field
 import os
 from pathlib import Path
 from enum import Enum
-from encoder import inference as encoder
+from models.encoder import inference as encoder
 import librosa
 from scipy.io.wavfile import write
 import re
 import numpy as np
-from mkgui.base.components.types import FileContent
-from vocoder.hifigan import inference as gan_vocoder
-from synthesizer.inference import Synthesizer
+from control.mkgui.base.components.types import FileContent
+from models.vocoder.hifigan import inference as gan_vocoder
+from models.synthesizer.inference import Synthesizer
 from typing import Any, Tuple
 import matplotlib.pyplot as plt
 
 # Constants
-AUDIO_SAMPLES_DIR = f"samples{os.sep}"
-SYN_MODELS_DIRT = f"synthesizer{os.sep}saved_models"
-ENC_MODELS_DIRT = f"encoder{os.sep}saved_models"
-VOC_MODELS_DIRT = f"vocoder{os.sep}saved_models"
+AUDIO_SAMPLES_DIR = f"data{os.sep}samples{os.sep}"
+SYN_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}synthesizer"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+VOC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}vocoder"
 TEMP_SOURCE_AUDIO = f"wavs{os.sep}temp_source.wav"
 TEMP_RESULT_AUDIO = f"wavs{os.sep}temp_result.wav"
 if not os.path.isdir("wavs"):
@@ -31,7 +31,7 @@ if os.path.isdir(SYN_MODELS_DIRT):
     synthesizers =  Enum('synthesizers', list((file.name, file) for file in Path(SYN_MODELS_DIRT).glob("**/*.pt")))
     print("Loaded synthesizer models: " + str(len(synthesizers)))
 else:
-    raise Exception(f"Model folder {SYN_MODELS_DIRT} doesn't exist.")
+    raise Exception(f"Model folder {SYN_MODELS_DIRT} doesn't exist. 请将模型文件位置移动到上述位置中进行重试！")
 
 if os.path.isdir(ENC_MODELS_DIRT):    
     encoders =  Enum('encoders', list((file.name, file) for file in Path(ENC_MODELS_DIRT).glob("**/*.pt")))
@@ -46,15 +46,16 @@ else:
     raise Exception(f"Model folder {VOC_MODELS_DIRT} doesn't exist.")
 
 
-
 class Input(BaseModel):
     message: str = Field(
         ..., example="欢迎使用工具箱, 现已支持中文输入！", alias="文本内容"
     )
     local_audio_file: audio_input_selection = Field(
-        ..., alias="输入语音（本地wav）",
+        ..., alias="选择语音（本地wav）",
         description="选择本地语音文件."
     )
+    record_audio_file: FileContent = Field(default=None, alias="录制语音",
+        description="录音.", is_recorder=True, mime_type="audio/wav")
     upload_audio_file: FileContent = Field(default=None, alias="或上传语音",
         description="拖拽或点击上传.", mime_type="audio/wav")
     encoder: encoders = Field(
@@ -104,7 +105,12 @@ def synthesize(input: Input) -> Output:
     gan_vocoder.load_model(Path(input.vocoder.value))
 
     # load file
-    if input.upload_audio_file != None:
+    if input.record_audio_file != None:
+        with open(TEMP_SOURCE_AUDIO, "w+b") as f:
+            f.write(input.record_audio_file.as_bytes())
+            f.seek(0)
+        wav, sample_rate = librosa.load(TEMP_SOURCE_AUDIO)
+    elif input.upload_audio_file != None:
         with open(TEMP_SOURCE_AUDIO, "w+b") as f:
             f.write(input.upload_audio_file.as_bytes())
             f.seek(0)

control/mkgui/app_vc.py

@@ -1,27 +1,26 @@
-from synthesizer.inference import Synthesizer
-from pydantic import BaseModel, Field
-from encoder import inference as speacker_encoder
-import torch
 import os
-from pathlib import Path
 from enum import Enum
-import ppg_extractor as Extractor
-import ppg2mel as Convertor
-import librosa
-from scipy.io.wavfile import write
-import re
-import numpy as np
-from mkgui.base.components.types import FileContent
-from vocoder.hifigan import inference as gan_vocoder
+from pathlib import Path
 from typing import Any, Tuple
-import matplotlib.pyplot as plt
 
+import librosa
+import matplotlib.pyplot as plt
+import torch
+from pydantic import BaseModel, Field
+from scipy.io.wavfile import write
+
+import models.ppg2mel as Convertor
+import models.ppg_extractor as Extractor
+from control.mkgui.base.components.types import FileContent
+from models.encoder import inference as speacker_encoder
+from models.synthesizer.inference import Synthesizer
+from models.vocoder.hifigan import inference as gan_vocoder
 
 # Constants
-AUDIO_SAMPLES_DIR = f'samples{os.sep}'
-EXT_MODELS_DIRT = f'ppg_extractor{os.sep}saved_models'
-CONV_MODELS_DIRT = f'ppg2mel{os.sep}saved_models'
-VOC_MODELS_DIRT = f'vocoder{os.sep}saved_models'
+AUDIO_SAMPLES_DIR = f'data{os.sep}samples{os.sep}'
+EXT_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg_extractor'
+CONV_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg2mel'
+VOC_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}vocoder'
 TEMP_SOURCE_AUDIO = f'wavs{os.sep}temp_source.wav'
 TEMP_TARGET_AUDIO = f'wavs{os.sep}temp_target.wav'
 TEMP_RESULT_AUDIO = f'wavs{os.sep}temp_result.wav'
@@ -132,9 +131,10 @@ def convert(input: Input) -> Output:
 
     ppg = extractor.extract_from_wav(src_wav)
     # Import necessary dependency of Voice Conversion
-    from utils.f0_utils import compute_f0, f02lf0, compute_mean_std, get_converted_lf0uv   
+    from utils.f0_utils import (compute_f0, compute_mean_std, f02lf0,
+                                get_converted_lf0uv)   
     ref_lf0_mean, ref_lf0_std = compute_mean_std(f02lf0(compute_f0(ref_wav)))
-    speacker_encoder.load_model(Path(f"encoder{os.sep}saved_models{os.sep}pretrained_bak_5805000.pt"))
+    speacker_encoder.load_model(Path(f"data{os.sep}ckpt{os.sep}encoder{os.sep}pretrained_bak_5805000.pt"))
     embed = speacker_encoder.embed_utterance(ref_wav)
     lf0_uv = get_converted_lf0uv(src_wav, ref_lf0_mean, ref_lf0_std, convert=True)
     min_len = min(ppg.shape[1], len(lf0_uv))

control/mkgui/base/ui/schema_utils.py

@@ -37,6 +37,12 @@ def is_single_file_property(property: Dict) -> bool:
     # TODO: binary?
     return property.get("format") == "byte"
 
+def is_single_autio_property(property: Dict) -> bool:
+    if property.get("type") != "string":
+        return False
+    # TODO: binary?
+    return property.get("format") == "bytes"
+
 
 def is_single_directory_property(property: Dict) -> bool:
     if property.get("type") != "string":

control/mkgui/base/ui/streamlit_ui.py

@@ -14,14 +14,13 @@ from fastapi.encoders import jsonable_encoder
 from loguru import logger
 from pydantic import BaseModel, ValidationError, parse_obj_as
 
-from mkgui.base import Opyrator
-from mkgui.base.core import name_to_title
-from mkgui.base.ui import schema_utils
-from mkgui.base.ui.streamlit_utils import CUSTOM_STREAMLIT_CSS
+from control.mkgui.base import Opyrator
+from control.mkgui.base.core import name_to_title
+from . import schema_utils
+from .streamlit_utils import CUSTOM_STREAMLIT_CSS
 
 STREAMLIT_RUNNER_SNIPPET = """
-from mkgui.base.ui import render_streamlit_ui
-from mkgui.base import Opyrator
+from control.mkgui.base.ui import render_streamlit_ui
 
 import streamlit as st
 
@@ -243,7 +242,14 @@ class InputUI:
         file_extension = None
         if "mime_type" in property:
             file_extension = mimetypes.guess_extension(property["mime_type"])
-
+        
+        if "is_recorder" in property:
+            from audio_recorder_streamlit import audio_recorder
+            audio_bytes = audio_recorder()
+            if audio_bytes:
+                streamlit_app.audio(audio_bytes, format="audio/wav")
+            return audio_bytes
+        
         uploaded_file = streamlit_app.file_uploader(
             **streamlit_kwargs, accept_multiple_files=False, type=file_extension
         )
@@ -263,6 +269,39 @@ class InputUI:
                 streamlit_app.video(bytes, format=property.get("mime_type"))
         return bytes
 
+    def _render_single_audio_input(
+            self, streamlit_app: st, key: str, property: Dict
+        ) -> Any:
+            # streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+            from audio_recorder_streamlit import audio_recorder
+            audio_bytes = audio_recorder()
+            if audio_bytes:
+                streamlit_app.audio(audio_bytes, format="audio/wav")
+            return audio_bytes
+
+            # file_extension = None
+            # if "mime_type" in property:
+            #     file_extension = mimetypes.guess_extension(property["mime_type"])
+
+            # uploaded_file = streamlit_app.file_uploader(
+            #     **streamlit_kwargs, accept_multiple_files=False, type=file_extension
+            # )
+            # if uploaded_file is None:
+            #     return None
+
+            # bytes = uploaded_file.getvalue()
+            # if property.get("mime_type"):
+            #     if is_compatible_audio(property["mime_type"]):
+            #         # Show audio
+            #         streamlit_app.audio(bytes, format=property.get("mime_type"))
+            #     if is_compatible_image(property["mime_type"]):
+            #         # Show image
+            #         streamlit_app.image(bytes)
+            #     if is_compatible_video(property["mime_type"]):
+            #         # Show video
+            #         streamlit_app.video(bytes, format=property.get("mime_type"))
+            # return bytes
+
     def _render_single_string_input(
         self, streamlit_app: st, key: str, property: Dict
     ) -> Any:
@@ -807,21 +846,20 @@ class OutputUI:
 
 def getOpyrator(mode: str) -> Opyrator:
     if mode == None or mode.startswith('VC'):
-        from mkgui.app_vc import convert
+        from control.mkgui.app_vc import convert
         return  Opyrator(convert)
     if mode == None or mode.startswith('预处理'):
-        from mkgui.preprocess import preprocess
+        from control.mkgui.preprocess import preprocess
         return  Opyrator(preprocess)
     if mode == None or mode.startswith('模型训练'):
-        from mkgui.train import train
+        from control.mkgui.train import train
         return  Opyrator(train)
     if mode == None or mode.startswith('模型训练(VC)'):
-        from mkgui.train_vc import train_vc
+        from control.mkgui.train_vc import train_vc
         return  Opyrator(train_vc)
-    from mkgui.app import synthesize
+    from control.mkgui.app import synthesize
     return Opyrator(synthesize)
     
-
 def render_streamlit_ui() -> None:
     # init
     session_state = st.session_state
@@ -845,7 +883,7 @@ def render_streamlit_ui() -> None:
     col2.title(title)
     col2.markdown("欢迎使用MockingBird Web 2")
 
-    image = Image.open(path.join('mkgui', 'static', 'mb.png'))
+    image = Image.open(path.join('control','mkgui', 'static', 'mb.png'))
     col1.image(image)
 
     st.markdown("---")
@@ -853,6 +891,13 @@ def render_streamlit_ui() -> None:
 
     with left:
         st.header("Control 控制")
+        # if session_state.mode in ["AI拟音", "VC拟音"] :
+            # from audiorecorder import audiorecorder
+            # audio = audiorecorder("Click to record", "Recording...")
+            # if len(audio) > 0:
+            #     # To play audio in frontend:
+            #     st.audio(audio.tobytes())
+            
         InputUI(session_state=session_state, input_class=opyrator.input_type).render_ui(st)
         execute_selected = st.button(opyrator.action)
         if execute_selected:

control/mkgui/preprocess.py

@@ -6,8 +6,8 @@ from typing import Any, Tuple
 
 
 # Constants
-EXT_MODELS_DIRT = f"ppg_extractor{os.sep}saved_models"
-ENC_MODELS_DIRT = f"encoder{os.sep}saved_models"
+EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
 
 
 if os.path.isdir(EXT_MODELS_DIRT):    
@@ -83,7 +83,7 @@ def preprocess(input: Input) -> Output:
     """Preprocess(预处理)"""
     finished = 0
     if input.model == Model.VC_PPG2MEL:
-        from ppg2mel.preprocess import preprocess_dataset
+        from models.ppg2mel.preprocess import preprocess_dataset
         finished = preprocess_dataset(
             datasets_root=Path(input.datasets_root),
             dataset=input.dataset,

control/mkgui/train.py

@@ -3,17 +3,17 @@ import os
 from pathlib import Path
 from enum import Enum
 from typing import Any
-from synthesizer.hparams import hparams
-from synthesizer.train import train as synt_train
+from models.synthesizer.hparams import hparams
+from models.synthesizer.train import train as synt_train
 
 # Constants
-SYN_MODELS_DIRT = f"synthesizer{os.sep}saved_models"
-ENC_MODELS_DIRT = f"encoder{os.sep}saved_models"
+SYN_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}synthesizer"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
 
 
-# EXT_MODELS_DIRT = f"ppg_extractor{os.sep}saved_models"
-# CONV_MODELS_DIRT = f"ppg2mel{os.sep}saved_models"
-# ENC_MODELS_DIRT = f"encoder{os.sep}saved_models"
+# EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+# CONV_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg2mel"
+# ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
 
 # Pre-Load models
 if os.path.isdir(SYN_MODELS_DIRT):    
@@ -96,7 +96,7 @@ def train(input: Input) -> Output:
     synt_train(
         input.run_id, 
         input.input_root, 
-        f"synthesizer{os.sep}saved_models", 
+        f"data{os.sep}ckpt{os.sep}synthesizer", 
         input.save_every, 
         input.backup_every, 
         input.log_every, 

control/mkgui/train_vc.py

@@ -4,14 +4,14 @@ from pathlib import Path
 from enum import Enum
 from typing import Any, Tuple
 import numpy as np
-from utils.load_yaml import HpsYaml
+from utils.hparams import HpsYaml
 from utils.util import AttrDict
 import torch
 
 # Constants
-EXT_MODELS_DIRT = f"ppg_extractor{os.sep}saved_models"
-CONV_MODELS_DIRT = f"ppg2mel{os.sep}saved_models"
-ENC_MODELS_DIRT = f"encoder{os.sep}saved_models"
+EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+CONV_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg2mel"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
 
 
 if os.path.isdir(EXT_MODELS_DIRT):    
@@ -144,7 +144,7 @@ def train_vc(input: Input) -> Output:
     if torch.cuda.is_available():
         torch.cuda.manual_seed_all(input.seed)
     mode = "train"
-    from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+    from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
     solver = Solver(config, params, mode)
     solver.load_data()
     solver.set_model()

control/toolbox/__init__.py

@@ -1,12 +1,12 @@
-from toolbox.ui import UI
-from encoder import inference as encoder
-from synthesizer.inference import Synthesizer
-from vocoder.wavernn import inference as rnn_vocoder
-from vocoder.hifigan import inference as gan_vocoder
-from vocoder.fregan import inference as fgan_vocoder
+from control.toolbox.ui import UI
+from models.encoder import inference as encoder
+from models.synthesizer.inference import Synthesizer
+from models.vocoder.wavernn import inference as rnn_vocoder
+from models.vocoder.hifigan import inference as gan_vocoder
+from models.vocoder.fregan import inference as fgan_vocoder
 from pathlib import Path
 from time import perf_counter as timer
-from toolbox.utterance import Utterance
+from control.toolbox.utterance import Utterance
 import numpy as np
 import traceback
 import sys
@@ -38,7 +38,8 @@ recognized_datasets = [
     "VoxCeleb2/dev/aac",
     "VoxCeleb2/test/aac",
     "VCTK-Corpus/wav48",
-    "aidatatang_200zh/corpus",
+    "aidatatang_200zh/corpus/test",
+    "aidatatang_200zh/corpus/train",
     "aishell3/test/wav",
     "magicdata/train",
 ]
@@ -396,7 +397,7 @@ class Toolbox:
         self.ui.log("Loading the extractor %s... " % model_fpath)
         self.ui.set_loading(1)
         start = timer()
-        import ppg_extractor as extractor
+        import models.ppg_extractor as extractor
         self.extractor = extractor.load_model(model_fpath)
         self.ui.log("Done (%dms)." % int(1000 * (timer() - start)), "append")
         self.ui.set_loading(0)
@@ -408,7 +409,7 @@ class Toolbox:
         self.ui.log("Loading the convertor %s... " % model_fpath)
         self.ui.set_loading(1)
         start = timer()
-        import ppg2mel as convertor
+        import models.ppg2mel as convertor
         self.convertor = convertor.load_model( model_fpath)
         self.ui.log("Done (%dms)." % int(1000 * (timer() - start)), "append")
         self.ui.set_loading(0)

control/toolbox/ui.py

@@ -3,9 +3,8 @@ from PyQt5 import QtGui
 from PyQt5.QtWidgets import *
 import matplotlib.pyplot as plt
 from matplotlib.backends.backend_qt5agg import FigureCanvasQTAgg as FigureCanvas
-from matplotlib.figure import Figure
-from encoder.inference import plot_embedding_as_heatmap
-from toolbox.utterance import Utterance
+from models.encoder.inference import plot_embedding_as_heatmap
+from control.toolbox.utterance import Utterance
 from pathlib import Path
 from typing import List, Set
 import sounddevice as sd

demo_toolbox.py

@@ -1,5 +1,5 @@
 from pathlib import Path
-from toolbox import Toolbox
+from control.toolbox import Toolbox
 from utils.argutils import print_args
 from utils.modelutils import check_model_paths
 import argparse
@@ -17,15 +17,15 @@ if __name__ == '__main__':
         "supported datasets.", default=None)
     parser.add_argument("-vc", "--vc_mode", action="store_true", 
                         help="Voice Conversion Mode(PPG based)")
-    parser.add_argument("-e", "--enc_models_dir", type=Path, default="encoder/saved_models", 
+    parser.add_argument("-e", "--enc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}encoder", 
                         help="Directory containing saved encoder models")
-    parser.add_argument("-s", "--syn_models_dir", type=Path, default="synthesizer/saved_models", 
+    parser.add_argument("-s", "--syn_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}synthesizer", 
                         help="Directory containing saved synthesizer models")
-    parser.add_argument("-v", "--voc_models_dir", type=Path, default="vocoder/saved_models", 
+    parser.add_argument("-v", "--voc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}vocoder", 
                         help="Directory containing saved vocoder models")
-    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default="ppg_extractor/saved_models", 
+    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg_extractor", 
                         help="Directory containing saved extrator models")
-    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default="ppg2mel/saved_models", 
+    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg2mel", 
                         help="Directory containing saved convert models")
     parser.add_argument("--cpu", action="store_true", help=\
         "If True, processing is done on CPU, even when a GPU is available.")

encoder/data_objects/__init__.py

@@ -1,2 +0,0 @@
-from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
-from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataLoader

gen_voice.py

@@ -1,23 +1,15 @@
-from encoder.params_model import model_embedding_size as speaker_embedding_size
-from utils.argutils import print_args
-from utils.modelutils import check_model_paths
-from synthesizer.inference import Synthesizer
-from encoder import inference as encoder
-from vocoder.wavernn import inference as rnn_vocoder
-from vocoder.hifigan import inference as gan_vocoder
+from models.synthesizer.inference import Synthesizer
+from models.encoder import inference as encoder
+from models.vocoder.hifigan import inference as gan_vocoder
 from pathlib import Path
 import numpy as np
 import soundfile as sf
-import librosa
-import argparse
 import torch
 import sys
 import os
 import re
 import cn2an
-import glob
 
-from audioread.exceptions import NoBackendError
 vocoder = gan_vocoder
 
 def gen_one_wav(synthesizer, in_fpath, embed, texts, file_name, seq):

models/encoder/audio.py

@@ -1,5 +1,5 @@
 from scipy.ndimage.morphology import binary_dilation
-from encoder.params_data import *
+from models.encoder.params_data import *
 from pathlib import Path
 from typing import Optional, Union
 from warnings import warn

models/encoder/data_objects/__init__.py

@@ -0,0 +1,2 @@
+from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
+from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataLoader

models/encoder/data_objects/speaker.py

@@ -1,5 +1,5 @@
-from encoder.data_objects.random_cycler import RandomCycler
-from encoder.data_objects.utterance import Utterance
+from models.encoder.data_objects.random_cycler import RandomCycler
+from models.encoder.data_objects.utterance import Utterance
 from pathlib import Path
 
 # Contains the set of utterances of a single speaker

models/encoder/data_objects/speaker_batch.py

@@ -1,6 +1,6 @@
 import numpy as np
 from typing import List
-from encoder.data_objects.speaker import Speaker
+from models.encoder.data_objects.speaker import Speaker
 
 class SpeakerBatch:
     def __init__(self, speakers: List[Speaker], utterances_per_speaker: int, n_frames: int):

models/encoder/data_objects/speaker_verification_dataset.py

@@ -1,7 +1,7 @@
-from encoder.data_objects.random_cycler import RandomCycler
-from encoder.data_objects.speaker_batch import SpeakerBatch
-from encoder.data_objects.speaker import Speaker
-from encoder.params_data import partials_n_frames
+from models.encoder.data_objects.random_cycler import RandomCycler
+from models.encoder.data_objects.speaker_batch import SpeakerBatch
+from models.encoder.data_objects.speaker import Speaker
+from models.encoder.params_data import partials_n_frames
 from torch.utils.data import Dataset, DataLoader
 from pathlib import Path
 

models/encoder/inference.py

@@ -1,8 +1,8 @@
-from encoder.params_data import *
-from encoder.model import SpeakerEncoder
-from encoder.audio import preprocess_wav   # We want to expose this function from here
+from models.encoder.params_data import *
+from models.encoder.model import SpeakerEncoder
+from models.encoder.audio import preprocess_wav   # We want to expose this function from here
 from matplotlib import cm
-from encoder import audio
+from models.encoder import audio
 from pathlib import Path
 import matplotlib.pyplot as plt
 import numpy as np

models/encoder/model.py

@@ -1,5 +1,5 @@
-from encoder.params_model import *
-from encoder.params_data import *
+from models.encoder.params_model import *
+from models.encoder.params_data import *
 from scipy.interpolate import interp1d
 from sklearn.metrics import roc_curve
 from torch.nn.utils import clip_grad_norm_

models/encoder/preprocess.py

@@ -1,8 +1,8 @@
 from multiprocess.pool import ThreadPool
-from encoder.params_data import *
-from encoder.config import librispeech_datasets, anglophone_nationalites
+from models.encoder.params_data import *
+from models.encoder.config import librispeech_datasets, anglophone_nationalites
 from datetime import datetime
-from encoder import audio
+from models.encoder import audio
 from pathlib import Path
 from tqdm import tqdm
 import numpy as np
@@ -22,7 +22,7 @@ class DatasetLog:
         self._log_params()
         
     def _log_params(self):
-        from encoder import params_data
+        from models.encoder import params_data
         self.write_line("Parameter values:")
         for param_name in (p for p in dir(params_data) if not p.startswith("__")):
             value = getattr(params_data, param_name)

models/encoder/train.py

@@ -1,7 +1,7 @@
-from encoder.visualizations import Visualizations
-from encoder.data_objects import SpeakerVerificationDataLoader, SpeakerVerificationDataset
-from encoder.params_model import *
-from encoder.model import SpeakerEncoder
+from models.encoder.visualizations import Visualizations
+from models.encoder.data_objects import SpeakerVerificationDataLoader, SpeakerVerificationDataset
+from models.encoder.params_model import *
+from models.encoder.model import SpeakerEncoder
 from utils.profiler import Profiler
 from pathlib import Path
 import torch

models/encoder/visualizations.py

@@ -1,4 +1,4 @@
-from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
+from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
 from datetime import datetime
 from time import perf_counter as timer
 import matplotlib.pyplot as plt
@@ -65,8 +65,8 @@ class Visualizations:
     def log_params(self):
         if self.disabled:
             return 
-        from encoder import params_data
-        from encoder import params_model
+        from models.encoder import params_data
+        from models.encoder import params_model
         param_string = "<b>Model parameters</b>:<br>"
         for param_name in (p for p in dir(params_model) if not p.startswith("__")):
             value = getattr(params_model, param_name)

models/ppg2mel/__init__.py

@@ -15,7 +15,7 @@ from .rnn_decoder_mol import Decoder
 from .utils.cnn_postnet import Postnet
 from .utils.vc_utils import get_mask_from_lengths
 
-from utils.load_yaml import HpsYaml
+from utils.hparams import HpsYaml
 
 class MelDecoderMOLv2(AbsMelDecoder):
     """Use an encoder to preprocess ppg."""

models/ppg2mel/preprocess.py

@@ -7,10 +7,10 @@ from pathlib import Path
 import soundfile
 import resampy
 
-from ppg_extractor import load_model
+from models.ppg_extractor import load_model
 import encoder.inference as Encoder
-from encoder.audio import preprocess_wav
-from encoder import audio
+from models.encoder.audio import preprocess_wav
+from models.encoder import audio
 from utils.f0_utils import compute_f0
 
 from torch.multiprocessing import Pool, cpu_count

models/ppg2mel/train.py

@@ -2,8 +2,8 @@ import sys
 import torch
 import argparse
 import numpy as np
-from utils.load_yaml import HpsYaml
-from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+from utils.hparams import HpsYaml
+from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
 
 # For reproducibility, comment these may speed up training
 torch.backends.cudnn.deterministic = True

models/ppg2mel/train/solver.py

@@ -8,7 +8,6 @@ from torch.utils.tensorboard import SummaryWriter
 
 from .option import default_hparas
 from utils.util import human_format, Timer
-from utils.load_yaml import HpsYaml
 
 
 class BaseSolver():

models/ppg2mel/train/train_linglf02mel_seq2seq_oneshotvc.py

@@ -14,7 +14,7 @@ from utils.data_load import OneshotVcDataset, MultiSpkVcCollate
 from .loss import MaskedMSELoss
 from .optim import Optimizer
 from utils.util import human_format
-from ppg2mel import MelDecoderMOLv2
+from models.ppg2mel import MelDecoderMOLv2
 
 
 class Solver(BaseSolver):

models/synthesizer/hparams.py

@@ -1,36 +1,4 @@
-import ast
-import pprint
-import json
-
-class HParams(object):
-    def __init__(self, **kwargs): self.__dict__.update(kwargs)
-    def __setitem__(self, key, value): setattr(self, key, value)
-    def __getitem__(self, key): return getattr(self, key)
-    def __repr__(self): return pprint.pformat(self.__dict__)
-
-    def parse(self, string):
-        # Overrides hparams from a comma-separated string of name=value pairs
-        if len(string) > 0:
-            overrides = [s.split("=") for s in string.split(",")]
-            keys, values = zip(*overrides)
-            keys = list(map(str.strip, keys))
-            values = list(map(str.strip, values))
-            for k in keys:
-                self.__dict__[k] = ast.literal_eval(values[keys.index(k)])
-        return self
-
-    def loadJson(self, dict):
-        print("\Loading the json with %s\n", dict)
-        for k in dict.keys():
-            if k not in ["tts_schedule", "tts_finetune_layers"]: 
-                self.__dict__[k] = dict[k]
-        return self
-
-    def dumpJson(self, fp):
-        print("\Saving the json with %s\n", fp)
-        with fp.open("w", encoding="utf-8") as f:
-            json.dump(self.__dict__, f)
-        return self
+from utils.hparams import HParams
 
 hparams = HParams(
         ### Signal Processing (used in both synthesizer and vocoder)
@@ -104,7 +72,7 @@ hparams = HParams(
         ### SV2TTS
         speaker_embedding_size = 256,               # Dimension for the speaker embedding
         silence_min_duration_split = 0.4,           # Duration in seconds of a silence for an utterance to be split
-        utterance_min_duration = 1.6,               # Duration in seconds below which utterances are discarded
+        utterance_min_duration = 0.5,               # Duration in seconds below which utterances are discarded
         use_gst = True,                             # Whether to use global style token    
         use_ser_for_gst = True,                     # Whether to use speaker embedding referenced for global style token  
         )

models/synthesizer/inference.py

@@ -1,16 +1,15 @@
 import torch
-from synthesizer import audio
-from synthesizer.hparams import hparams
-from synthesizer.models.tacotron import Tacotron
-from synthesizer.utils.symbols import symbols
-from synthesizer.utils.text import text_to_sequence
-from vocoder.display import simple_table
+from models.synthesizer import audio
+from models.synthesizer.hparams import hparams
+from models.synthesizer.models.tacotron import Tacotron
+from models.synthesizer.utils.symbols import symbols
+from models.synthesizer.utils.text import text_to_sequence
+from models.vocoder.display import simple_table
 from pathlib import Path
 from typing import Union, List
 import numpy as np
 import librosa
 from utils import logmmse
-import json
 from pypinyin import lazy_pinyin, Style
 
 class Synthesizer:
@@ -48,8 +47,7 @@ class Synthesizer:
         # Try to scan config file
         model_config_fpaths = list(self.model_fpath.parent.rglob("*.json"))
         if len(model_config_fpaths)>0 and model_config_fpaths[0].exists():
-            with model_config_fpaths[0].open("r", encoding="utf-8") as f:
-                hparams.loadJson(json.load(f))
+            hparams.loadJson(model_config_fpaths[0])
         """
         Instantiates and loads the model given the weights file that was passed in the constructor.
         """