From 5a6a594735b5b041c768f28e7e5fc709a6b77e55 Mon Sep 17 00:00:00 2001
From: litagin02 <litagin02@gmail.com>
Date: Tue, 2 Jan 2024 19:30:59 +0900
Subject: [PATCH] Update ui

---
 webui_dataset.py       | 15 +++++++++++----
 webui_style_vectors.py | 11 ++++-------
 2 files changed, 15 insertions(+), 11 deletions(-)

diff --git a/webui_dataset.py b/webui_dataset.py
index e65c356..3aeb765 100644
--- a/webui_dataset.py
+++ b/webui_dataset.py
@@ -27,6 +27,9 @@ def do_slice(model_name: str, min_sec: float, max_sec: float, input_dir="inputs"
 
 
 def do_transcribe(model_name, whisper_model, compute_type, language, initial_prompt):
+    if initial_prompt == "":
+        initial_prompt = "こんにちは。元気、ですかー？私は……ちゃんと元気だよ！"
+    logger.debug(f"initial_prompt: {initial_prompt}")
     input_dir = os.path.join("Data", model_name, "raw")
     output_file = os.path.join("Data", model_name, "esd.list")
     result = run_script_with_log(
@@ -45,7 +48,7 @@ def do_transcribe(model_name, whisper_model, compute_type, language, initial_pro
             "--language",
             language,
             "--initial_prompt",
-            initial_prompt,
+            '"initial_prompt"',
         ]
     )
     return "音声の文字起こしが完了しました。"
@@ -89,8 +92,9 @@ with gr.Blocks(theme="NoCrypt/miku") as app:
             result1 = gr.Textbox(label="結果")
     with gr.Row():
         with gr.Column():
-            whisper_model = gr.Radio(
+            whisper_model = gr.Dropdown(
                 ["tiny", "base", "small", "medium", "large", "large-v2", "large-v3"],
+                label="Whisperモデル",
                 value="large-v3",
             )
             compute_type = gr.Dropdown(
@@ -104,10 +108,13 @@ with gr.Blocks(theme="NoCrypt/miku") as app:
                     "bfloat16",
                     "float32",
                 ],
+                label="計算精度",
                 value="bfloat16",
             )
-            language = gr.Dropdown(["ja", "en", "zh"], value="ja")
-            initial_prompt = gr.Textbox(label="初期プロンプトを入力してください（省略可）")
+            language = gr.Dropdown(["ja", "en", "zh"], value="ja", label="言語")
+            initial_prompt = gr.Textbox(
+                label="初期プロンプト（日本語の場合は省略可）", placeholder="こんにちは。元気、ですかー？私は……ちゃんと元気だよ！"
+            )
         transcribe_button = gr.Button("音声の文字起こし")
         result2 = gr.Textbox(label="結果")
     slice_button.click(
diff --git a/webui_style_vectors.py b/webui_style_vectors.py
index 7635a6b..4f06020 100644
--- a/webui_style_vectors.py
+++ b/webui_style_vectors.py
@@ -24,12 +24,11 @@ centroids = []
 def load(model_name):
     global wav_files, x, x_tsne, mean
     wavs_dir = os.path.join("Data", model_name, "wavs")
-    wav_files = [
-        os.path.join(wavs_dir, wav_path)
-        for wav_path in os.listdir(wavs_dir)
-        if wav_path.endswith(".wav")
+    style_vector_files = [
+        os.path.join(wavs_dir, f) for f in os.listdir(wavs_dir) if f.endswith(".npy")
     ]
-    style_vectors = [np.load(f"{wav_path}.npy") for wav_path in wav_files]
+    wav_files = [f.replace(".npy", "") for f in style_vector_files]
+    style_vectors = [np.load(f) for f in style_vector_files]
     x = np.array(style_vectors)
     mean = np.mean(x, axis=0)
 
@@ -166,8 +165,6 @@ def save_style_vectors_from_files(model_name, audio_files_text, style_names_text
         style_vectors.append(np.load(f"{path}.npy"))
     style_vectors = np.stack(style_vectors)
     style_vector_path = os.path.join(result_dir, "style_vectors.npy")
-    if os.path.exists(style_vector_path):
-        return f"{style_vector_path}が既に存在します。削除するか別の名前にバックアップしてください。"
     np.save(style_vector_path, style_vectors)
 
     # config.jsonの更新