Update app.py

app.py

@@ -1,7 +1,7 @@
 """
-MediaTek BreezyVoice 混合版本
-HF 下載模型 + GitHub clone 推論程式碼
-尊重作者設計，同時在 HF 上顯示模型引用
+MediaTek BreezyVoice Hybrid Version
+HF downloads the model + GitHub clone for inference code
+Respect the author’s design and display model citation on HF
 """
 
 import gradio as gr
@@ -15,53 +15,53 @@ import subprocess
 import sys
 from pathlib import Path
 
-# 設置單線程模式避免多進程衝突
+# Set single-thread mode to avoid multiprocessing conflicts
 torch.set_num_threads(1)
 os.environ['OMP_NUM_THREADS'] = '1'
 os.environ['MKL_NUM_THREADS'] = '1'
 os.environ['TOKENIZERS_PARALLELISM'] = 'false'
 os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'expandable_segments:True'
 
-# 預設參考語音範例 (約20秒朗讀)
-DEFAULT_REFERENCE_TEXT = "台灣是個美麗的島嶼，擁有豐富的自然景觀和多元的文化特色。從北部的陽明山到南部的墾丁，每個地方都有獨特的魅力。四季分明的氣候讓這裡的生活充滿變化，春天櫻花盛開，夏天海灘戲水，秋天楓葉飄香，冬天溫泉暖身。"
+# Default reference voice sample (about 20 seconds of reading)
+DEFAULT_REFERENCE_TEXT = "Taiwan is a beautiful island with rich natural scenery and diverse cultural characteristics. From Yangmingshan in the north to Kenting in the south, each place has its own unique charm. The four distinct seasons make life here full of variety—cherry blossoms in spring, beach fun in summer, maple leaves in autumn, and hot springs in winter."
 
-# 全域變數
+# Global variables
 cosyvoice = None
 setup_completed = False
 
 @spaces.GPU(duration=300)
 def setup_breezyvoice_hybrid():
-    """設置 BreezyVoice 混合版本 - HF 模型 + GitHub 程式碼"""
+    """Setup BreezyVoice Hybrid Version - HF model + GitHub code"""
     global cosyvoice, setup_completed
     
     if setup_completed:
-        return "✅ BreezyVoice 混合版已準備就緒"
+        return "✅ BreezyVoice Hybrid version is ready"
     
     try:
-        print("🔧 正在設置 BreezyVoice 混合版本...")
-        print("📊 策略: HF 下載模型 + GitHub 下載推論程式碼")
+        print("🔧 Setting up BreezyVoice Hybrid Version...")
+        print("📊 Strategy: Download model from HF + Download inference code from GitHub")
         
-        # 1. 從 HuggingFace 下載模型 (尊重作者，顯示模型引用)
+        # 1. Download model from HuggingFace (respecting the author with citation)
         try:
             from huggingface_hub import snapshot_download
-            print("📥 從 HuggingFace 下載 MediaTek-Research/BreezyVoice 模型...")
+            print("📥 Downloading MediaTek-Research/BreezyVoice model from HuggingFace...")
             
             model_path = snapshot_download(
                 "MediaTek-Research/BreezyVoice", 
                 cache_dir="/tmp/hf_models",
                 local_files_only=False
             )
-            print(f"✅ HF 模型下載完成: {model_path}")
+            print(f"✅ HF model downloaded: {model_path}")
             
         except Exception as e:
-            print(f"⚠️ HF 模型下載失敗: {e}")
-            print("🔄 改用 GitHub 直接下載模型...")
-            model_path = "MediaTek-Research/BreezyVoice"  # 備用方案
+            print(f"⚠️ HF model download failed: {e}")
+            print("🔄 Falling back to direct GitHub model path...")
+            model_path = "MediaTek-Research/BreezyVoice"  # fallback
         
-        # 2. 從 GitHub 下載推論程式碼 (作者設計的使用方式)
+        # 2. Download inference code from GitHub (as designed by the author)
         repo_path = "/tmp/BreezyVoice"
         if not os.path.exists(repo_path):
-            print("📥 從 GitHub 下載 BreezyVoice 推論程式碼...")
+            print("📥 Downloading BreezyVoice inference code from GitHub...")
             result = subprocess.run([
                 "git", "clone", 
                 "https://github.com/mtkresearch/BreezyVoice.git",
@@ -69,48 +69,48 @@ def setup_breezyvoice_hybrid():
             ], capture_output=True, text=True, timeout=300)
             
             if result.returncode != 0:
-                raise Exception(f"程式碼下載失敗: {result.stderr}")
+                raise Exception(f"Code download failed: {result.stderr}")
         
-        # 3. 添加模組路徑
+        # 3. Add module path
         sys.path.insert(0, repo_path)
         
-        # 4. 導入作者設計的核心模組
+        # 4. Import core module from author
         try:
             from single_inference import CustomCosyVoice
-            print("✅ BreezyVoice 核心模組導入成功")
+            print("✅ BreezyVoice core module imported successfully")
         except ImportError as e:
-            raise Exception(f"模組導入失敗: {e}")
+            raise Exception(f"Module import failed: {e}")
         
-        # 5. 初始化模型 (混合方式：HF 模型路徑 + GitHub 程式碼)
-        print(f"🔄 初始化混合版本模型...")
-        print(f"📍 模型路徑: {model_path}")
+        # 5. Initialize model (hybrid: HF model path + GitHub code)
+        print(f"🔄 Initializing hybrid model...")
+        print(f"📍 Model path: {model_path}")
         
         cosyvoice = CustomCosyVoice(model_path)
         
         setup_completed = True
-        print("✅ BreezyVoice 混合版設置完成!")
+        print("✅ BreezyVoice hybrid setup complete!")
         
-        # 檢查 VRAM 使用
+        # Check VRAM usage
         if torch.cuda.is_available():
             vram_used = torch.cuda.memory_allocated() / 1024**3
-            return f"✅ BreezyVoice 混合版設置完成！\n📊 模型來源: HuggingFace MediaTek-Research/BreezyVoice\n🔧 推論程式: GitHub mtkresearch/BreezyVoice\n💾 VRAM 使用: {vram_used:.2f}GB"
+            return f"✅ BreezyVoice Hybrid setup complete!\n📊 Model source: HuggingFace MediaTek-Research/BreezyVoice\n🔧 Inference code: GitHub mtkresearch/BreezyVoice\n💾 VRAM usage: {vram_used:.2f}GB"
         
-        return "✅ BreezyVoice 混合版設置完成！"
+        return "✅ BreezyVoice Hybrid setup complete!"
         
     except Exception as e:
-        print(f"❌ 設置失敗: {str(e)}")
-        return f"❌ 設置失敗: {str(e)}"
+        print(f"❌ Setup failed: {str(e)}")
+        return f"❌ Setup failed: {str(e)}"
 
 @spaces.GPU(duration=180)
 def breezy_voice_clone_hybrid(speaker_audio, content_text, speaker_transcription=None):
-    """執行 BreezyVoice 語音克隆 - 混合版本"""
+    """Run BreezyVoice voice cloning - hybrid version"""
     global cosyvoice
     
     if speaker_audio is None:
-        return None, "❌ 請先上傳或錄製參考語音"
+        return None, "❌ Please upload or record a reference voice first"
     
     if not content_text.strip():
-        return None, "❌ 請輸入要合成的文字"
+        return None, "❌ Please enter the text to synthesize"
     
     if not setup_completed or cosyvoice is None:
         setup_status = setup_breezyvoice_hybrid()
@@ -119,118 +119,109 @@ def breezy_voice_clone_hybrid(speaker_audio, content_text, speaker_transcription
     
     try:
         with tempfile.TemporaryDirectory() as temp_dir:
-            # 處理輸入音訊
+            # Handle input audio
             input_audio_path = os.path.join(temp_dir, "speaker_voice.wav")
             output_audio_path = os.path.join(temp_dir, "cloned_voice.wav")
             
-            # 保存參考音訊
+            # Save reference audio
             sample_rate, audio_data = speaker_audio
             torchaudio.save(input_audio_path, torch.tensor(audio_data).unsqueeze(0), sample_rate)
             
-            # 使用參考轉錄或預設值
+            # Use provided or default transcription
             if not speaker_transcription or not speaker_transcription.strip():
                 speaker_transcription = DEFAULT_REFERENCE_TEXT
             
-            print(f"🎤 合成文字: {content_text}")
-            print(f"📝 參考轉錄: {speaker_transcription}")
+            print(f"🎤 Synthesizing text: {content_text}")
+            print(f"📝 Reference transcription: {speaker_transcription}")
             
-            # 執行語音合成 - 混合版本
+            # Run synthesis - hybrid version
             synthesis_start = time.time()
             
             try:
-                # 載入音訊為 16kHz
                 from cosyvoice.utils.file_utils import load_wav
                 prompt_speech_16k = load_wav(input_audio_path, 16000)
                 
-                print("🔄 執行混合版語音合成推論...")
+                print("🔄 Running hybrid voice synthesis inference...")
                 
-                # 使用作者設計的 no_normalize 版本 (與本地測試一致)
+                # Use no_normalize version (as in local testing)
                 output = cosyvoice.inference_zero_shot_no_normalize(
                     content_text, 
                     speaker_transcription, 
                     prompt_speech_16k
                 )
                 
-                # 保存輸出音訊
                 if output is not None and 'tts_speech' in output:
                     tts_speech = output['tts_speech']
                     torchaudio.save(output_audio_path, tts_speech, 22050)
                     
                 synthesis_time = time.time() - synthesis_start
                 
-                # 檢查輸出
                 if os.path.exists(output_audio_path):
-                    # 讀取合成的音訊
                     synthesized_audio, file_sample_rate = torchaudio.load(output_audio_path)
                     synthesized_audio = synthesized_audio.numpy()
                     
-                    # 計算音訊長度
                     audio_duration = synthesized_audio.shape[1] / file_sample_rate
                     rtf = synthesis_time / audio_duration if audio_duration > 0 else float('inf')
                     
-                    # 檢查 VRAM 使用
                     vram_info = ""
                     if torch.cuda.is_available():
                         vram_used = torch.cuda.memory_allocated() / 1024**3
                         vram_info = f"💾 VRAM: {vram_used:.2f}GB"
                     
-                    status = f"""✅ 混合版語音克隆成功！
+                    status = f"""✅ Hybrid voice cloning successful!
 
-🎙️ 參考語音: {len(audio_data)/sample_rate:.1f}秒
-📝 合成內容: {content_text}
-📝 使用轉錄: {speaker_transcription[:30]}...
-⏱️ 合成時間: {synthesis_time:.1f}秒
-🎵 輸出長度: {audio_duration:.1f}秒
-📊 RTF: {rtf:.3f} {'(實時)' if rtf < 1.0 else '(非實時)'}
+🎙️ Reference audio: {len(audio_data)/sample_rate:.1f}s
+📝 Synthesized content: {content_text}
+📝 Used transcription: {speaker_transcription[:30]}...
+⏱️ Synthesis time: {synthesis_time:.1f}s
+🎵 Output length: {audio_duration:.1f}s
+📊 RTF: {rtf:.3f} {'(real-time)' if rtf < 1.0 else '(non real-time)'}
 {vram_info}
-🤖 模型: MediaTek BreezyVoice 混合版
-📊 模型來源: HuggingFace MediaTek-Research/BreezyVoice
-🔧 推論程式: GitHub mtkresearch/BreezyVoice"""
+🤖 Model: MediaTek BreezyVoice Hybrid
+📊 Model source: HuggingFace MediaTek-Research/BreezyVoice
+🔧 Inference code: GitHub mtkresearch/BreezyVoice"""
                     
                     return (file_sample_rate, synthesized_audio[0]), status
                 else:
-                    return None, "❌ 語音合成失敗：未生成輸出檔案"
+                    return None, "❌ Synthesis failed: No output file generated"
                     
             except Exception as e:
                 import traceback
                 traceback.print_exc()
-                return None, f"❌ 語音合成失敗: {str(e)}"
+                return None, f"❌ Synthesis failed: {str(e)}"
                 
     except Exception as e:
         import traceback
         traceback.print_exc()
-        return None, f"❌ 處理錯誤: {str(e)}"
+        return None, f"❌ Processing error: {str(e)}"
 
 def load_example_text():
-    """載入預設範例文字"""
+    """Load default example text"""
     return DEFAULT_REFERENCE_TEXT
 
-# 創建 Gradio 界面
-with gr.Blocks(title="BreezyVoice 混合版", theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🎭 MediaTek BreezyVoice 混合版")
-    gr.Markdown("**零樣本語音克隆系統** - 專為台灣繁體中文優化 (混合版)")
-    gr.Markdown("📊 **技術架構**: HuggingFace 模型 + GitHub 推論程式碼")
+# Build Gradio interface
+with gr.Blocks(title="BreezyVoice Hybrid", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🎭 MediaTek BreezyVoice Hybrid")
+    gr.Markdown("**Zero-shot voice cloning system** — optimized for Traditional Chinese (Taiwan)")
+    gr.Markdown("📊 **Architecture**: HuggingFace model + GitHub inference code")
     
-    # 初始化狀態顯示
     setup_status = gr.Textbox(
-        label="🔧 系統狀態",
-        value="⏳ 準備初始化 BreezyVoice 混合版...",
+        label="🔧 System Status",
+        value="⏳ Preparing BreezyVoice Hybrid initialization...",
         interactive=False
     )
     
-    # 初始化按鈕
-    init_btn = gr.Button("🚀 初始化 BreezyVoice 混合版", variant="primary")
+    init_btn = gr.Button("🚀 Initialize BreezyVoice Hybrid", variant="primary")
     
     with gr.Row():
         with gr.Column(scale=1):
-            gr.Markdown("### 🎙️ 步驟 1: 上傳參考語音")
-            gr.Markdown("請照著下面的範例文字朗讀，上傳 5-20 秒清晰語音")
+            gr.Markdown("### 🎙️ Step 1: Upload Reference Voice")
+            gr.Markdown("Please read the following sample text aloud and upload a 5–20 second clear voice recording")
             
-            # 顯示範例文字
-            gr.Markdown("#### 📖 建議朗讀範例：")
+            gr.Markdown("#### 📖 Recommended Reading Sample:")
             example_display = gr.Textbox(
                 value=DEFAULT_REFERENCE_TEXT,
-                label="請照著這段文字朗讀 (約20秒)",
+                label="Read this text aloud (around 20 seconds)",
                 lines=4,
                 interactive=False
             )
@@ -238,90 +229,74 @@ with gr.Blocks(title="BreezyVoice 混合版", theme=gr.themes.Soft()) as demo:
             speaker_audio = gr.Audio(
                 sources=["microphone", "upload"],
                 type="numpy",
-                label="參考語音錄音 (照著上面文字念)"
+                label="Reference voice recording (read the text above)"
             )
             
-            gr.Markdown("### 📝 步驟 2: 輸入合成文字")
+            gr.Markdown("### 📝 Step 2: Enter Text to Synthesize")
             content_text = gr.Textbox(
                 lines=3,
-                placeholder="請輸入要用克隆聲音說出的內容...",
-                label="合成文字內容",
-                value="歡迎來到我們的語音合成��統！這個技術可以模仿任何人的聲音，讓文字轉換成自然流暢的語音。"
+                placeholder="Enter the text you want spoken in the cloned voice...",
+                label="Synthesis text",
+                value="Welcome to our voice synthesis system! This technology can mimic anyone’s voice and convert text into natural, fluent speech."
             )
             
-            gr.Markdown("### 🔤 步驟 3: 參考語音轉錄")
+            gr.Markdown("### 🔤 Step 3: Reference Voice Transcription")
             speaker_transcription = gr.Textbox(
                 lines=3,
-                label="參考語音轉錄 (預設範例)",
+                label="Reference transcription (default example)",
                 value=DEFAULT_REFERENCE_TEXT
             )
             
-            # 載入範例按鈕
-            load_example_btn = gr.Button("📄 載入預設範例", variant="secondary")
-            
-            clone_btn = gr.Button("🎭 開始混合版語音克隆", variant="primary", size="lg")
+            load_example_btn = gr.Button("📄 Load Default Example", variant="secondary")
+            clone_btn = gr.Button("🎭 Start Hybrid Voice Cloning", variant="primary", size="lg")
             
         with gr.Column(scale=1):
-            gr.Markdown("### 🎵 克隆結果")
+            gr.Markdown("### 🎵 Cloning Result")
             
             result_audio = gr.Audio(
-                label="克隆的語音",
+                label="Cloned Voice",
                 type="numpy"
             )
             
             result_status = gr.Textbox(
-                label="📋 處理狀態",
+                label="📋 Processing Status",
                 lines=15,
                 max_lines=20,
                 interactive=False
             )
     
-    # 使用說明
-    with gr.Accordion("📖 使用說明", open=False):
+    with gr.Accordion("📖 User Guide", open=False):
         gr.Markdown(f"""
-        ## 🎯 最佳使用方式
-        1. **📖 朗讀範例**: 請照著範例文字清晰朗讀
-        2. **🎙️ 錄音要求**: 5-20 秒，環境安靜，發音清楚  
-        3. **✨ 克隆效果**: 系統會用您的聲音說出任何文字
+        ## 🎯 Best Practices
+        1. **📖 Read the sample clearly**  
+        2. **🎙️ Recording requirements**: 5–20 seconds, quiet environment, clear pronunciation  
+        3. **✨ Cloning effect**: The system will use your voice to speak any text
         
-        ## 📝 範例文字內容
+        ## 📝 Sample Text
         ```
         {DEFAULT_REFERENCE_TEXT}
         ```
         
-        ## ⚡ 技術特色
-        - 🇹🇼 台灣繁體中文專門優化
-        - 🎯 零樣本克隆（無需訓練）
-        - ⚡ ZeroGPU 加速處理
-        - 🔊 MediaTek 先進語音合成技術
-        - 🤗 HuggingFace 模型引用 + GitHub 推論程式碼
+        ## ⚡ Technical Highlights
+        - 🇹🇼 Optimized for Traditional Chinese (Taiwan)
+        - 🎯 Zero-shot cloning (no training required)
+        - ⚡ ZeroGPU accelerated processing
+        - 🔊 MediaTek advanced speech synthesis
+        - 🤗 HuggingFace model citation + GitHub inference code
         
-        ## 💡 混合版本優勢
-        - **模型來源**: 直接引用 HuggingFace MediaTek-Research/BreezyVoice
-        - **推論程式**: 使用作者原始 GitHub mtkresearch/BreezyVoice
-        - **最佳實踐**: 尊重作者設計，同時展示模型使用
-        - **技術穩定**: 結合官方模型和原始推論程式碼
+        ## 💡 Advantages of the Hybrid Version
+        - **Model Source**: HuggingFace MediaTek-Research/BreezyVoice  
+        - **Inference Code**: GitHub mtkresearch/BreezyVoice  
+        - **Best Practice**: Respect author design and display model usage  
+        - **Stability**: Combines official model with original inference code
         
-        ## 🙏 致謝
-        感謝 MediaTek Research 團隊開發 BreezyVoice 模型
+        ## 🙏 Acknowledgment
+        Thanks to MediaTek Research for developing the BreezyVoice model.
         """)
     
-    # 事件綁定
-    init_btn.click(
-        fn=setup_breezyvoice_hybrid,
-        outputs=[setup_status]
-    )
-    
-    load_example_btn.click(
-        fn=load_example_text,
-        outputs=[speaker_transcription]
-    )
-    
-    clone_btn.click(
-        fn=breezy_voice_clone_hybrid,
-        inputs=[speaker_audio, content_text, speaker_transcription],
-        outputs=[result_audio, result_status]
-    )
+    init_btn.click(fn=setup_breezyvoice_hybrid, outputs=[setup_status])
+    load_example_btn.click(fn=load_example_text, outputs=[speaker_transcription])
+    clone_btn.click(fn=breezy_voice_clone_hybrid, inputs=[speaker_audio, content_text, speaker_transcription], outputs=[result_audio, result_status])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()