From 8d7d7c32652356021b70655f3ccc44463c5ccb47 Mon Sep 17 00:00:00 2001
From: litagin02 <litagin02@gmail.com>
Date: Sun, 25 Feb 2024 17:42:50 +0900
Subject: [PATCH] Avoid using pyworld by default

---
 app.py                            |  6 +++++-
 common/tts_model.py               | 24 ++++++++++++++++--------
 requirements.txt                  |  8 +++-----
 scripts/Update-to-Dict-Editor.bat |  4 +---
 4 files changed, 25 insertions(+), 17 deletions(-)

diff --git a/app.py b/app.py
index d99229b..e17a0a5 100644
--- a/app.py
+++ b/app.py
@@ -202,7 +202,9 @@ examples = [
 initial_md = f"""
 # Style-Bert-VITS2 ver {LATEST_VERSION} 音声合成
 
-注意: 初期からある[jvnvのモデル](https://huggingface.co/litagin/style_bert_vits2_jvnv)は、[JVNVコーパス（言語音声と非言語音声を持つ日本語感情音声コーパス）](https://sites.google.com/site/shinnosuketakamichi/research-topics/jvnv_corpus)で学習されたモデルです。ライセンスは[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/deed.ja)です。
+- Ver 2.3で追加されたエディターのほうが実際に読み上げさせるには使いやすいかもしれません。`Editor.bat`か`python server_editor.py`で起動できます。
+
+- 初期からある[jvnvのモデル](https://huggingface.co/litagin/style_bert_vits2_jvnv)は、[JVNVコーパス（言語音声と非言語音声を持つ日本語感情音声コーパス）](https://sites.google.com/site/shinnosuketakamichi/research-topics/jvnv_corpus)で学習されたモデルです。ライセンスは[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/deed.ja)です。
 """
 
 how_to_md = """
@@ -317,6 +319,7 @@ if __name__ == "__main__":
                     value=1,
                     step=0.05,
                     label="音程(1以外では音質劣化)",
+                    visible=False,  # pyworldが必要
                 )
                 intonation_scale = gr.Slider(
                     minimum=0,
@@ -324,6 +327,7 @@ if __name__ == "__main__":
                     value=1,
                     step=0.1,
                     label="抑揚(1以外では音質劣化)",
+                    visible=False,  # pyworldが必要
                 )
 
                 line_split = gr.Checkbox(
diff --git a/common/tts_model.py b/common/tts_model.py
index 4f3932a..de4e830 100644
--- a/common/tts_model.py
+++ b/common/tts_model.py
@@ -5,7 +5,7 @@ from typing import Optional, Union
 
 import gradio as gr
 import numpy as np
-import pyworld
+
 import torch
 from gradio.processing_utils import convert_to_16_bit_wav
 
@@ -33,6 +33,13 @@ def adjust_voice(fs, wave, pitch_scale, intonation_scale):
         # 初期値の場合は、音質劣化を避けるためにそのまま返す
         return fs, wave
 
+    try:
+        import pyworld
+    except ImportError:
+        raise ImportError(
+            "pyworld is not installed. Please install it by `pip install pyworld`"
+        )
+
     # pyworldでf0を加工して合成
     # pyworldよりもよいのがあるかもしれないが……
 
@@ -197,16 +204,17 @@ class Model:
                         audios.append(np.zeros(int(44100 * split_interval)))
                 audio = np.concatenate(audios)
         logger.info("Audio data generated successfully")
-        fs, audio = adjust_voice(
-            fs=self.hps.data.sampling_rate,
-            wave=audio,
-            pitch_scale=pitch_scale,
-            intonation_scale=intonation_scale,
-        )
+        if not (pitch_scale == 1.0 and intonation_scale == 1.0):
+            _, audio = adjust_voice(
+                fs=self.hps.data.sampling_rate,
+                wave=audio,
+                pitch_scale=pitch_scale,
+                intonation_scale=intonation_scale,
+            )
         with warnings.catch_warnings():
             warnings.simplefilter("ignore")
             audio = convert_to_16_bit_wav(audio)
-        return (fs, audio)
+        return (self.hps.data.sampling_rate, audio)
 
 
 class ModelHolder:
diff --git a/requirements.txt b/requirements.txt
index b346afc..6bf329f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,16 +1,14 @@
 cmudict
 cn2an
-faster-whisper>=0.10.0,<1.0.0
+faster-whisper==0.10.1
 g2p_en
 GPUtil
 gradio
-jaconv
 jieba
 langid
 librosa==0.9.2
 loguru
 matplotlib
-mecab-python3
 num2words
 numba
 numpy
@@ -20,12 +18,12 @@ pyloudnorm
 # pyopenjtalk-prebuilt  # Should be manually uninstalled
 pyopenjtalk-dict
 pypinyin
-pyworld
+# pyworld  # Not supported on Windows without Cython...
 PyYAML
 requests
 safetensors
 scipy
 tensorboard
-torch>=2.1,<2.2 # For users without GPU or colab
+torch>=2.1,<2.2
 transformers
 umap-learn
diff --git a/scripts/Update-to-Dict-Editor.bat b/scripts/Update-to-Dict-Editor.bat
index e10060d..5fde8fc 100644
--- a/scripts/Update-to-Dict-Editor.bat
+++ b/scripts/Update-to-Dict-Editor.bat
@@ -28,14 +28,12 @@ xcopy /QSY .\Style-Bert-VITS2-dev\ .\Style-Bert-VITS2\
 rmdir /s /q Style-Bert-VITS2-dev
 if %errorlevel% neq 0 ( pause & popd & exit /b %errorlevel% )
 
-@REM 仮想環境のpip requirements.txtを更新
+@REM 仮想環境を有効化
 
 echo call .\Style-Bert-VITS2\scripts\activate.bat
 call .\Style-Bert-VITS2\venv\Scripts\activate.bat
 if %errorlevel% neq 0 ( pause & popd & exit /b %errorlevel% )
 
-pause
-
 @REM pyopenjtalk-prebuiltやpyopenjtalkが入っていたら削除
 echo python -m pip uninstall -y pyopenjtalk-prebuilt pyopenjtalk
 python -m pip uninstall -y pyopenjtalk-prebuilt pyopenjtalk