feat: add vscode debugger

Reviewed-on: terminator/ai-voice-cloning-terminator#1

.vscode/launch.json

@@ -0,0 +1,16 @@
+{
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Python: Debug Main.py",
+            "type": "python",
+            "request": "launch",
+            "program": "${workspaceFolder}/src/main.py",
+            "console": "integratedTerminal",
+            "env": {
+                "PYTHONPATH": "${workspaceFolder}"
+            },
+            "preLaunchTask": "Setup Environment",
+        }
+    ]
+}

.vscode/tasks.json

@@ -0,0 +1,12 @@
+{
+  "version": "2.0.0",
+  "tasks": [
+      {
+          "label": "Setup Environment",
+          "type": "shell",
+          "command": "python",
+          "args": ["activate_env.py"],
+          "problemMatcher": []
+      }
+  ]
+}

activate_env.py

@@ -0,0 +1,9 @@
+import subprocess
+import os
+
+# Ativa o ambiente virtual
+subprocess.run([r'.\venv\Scripts\activate.bat'], shell=True)
+
+# Configura as variáveis de ambiente
+os.environ['PATH'] = r'.\bin;' + os.environ['PATH']
+os.environ['PYTHONUTF8'] = '1'

notebook_colab.ipynb

@@ -38,10 +38,24 @@
             
          ],
          "source":[
-            "!apt install python3.8-venv\n",
+            "!apt install python3.10-venv\n",
             "!git clone https://git.ecker.tech/mrq/ai-voice-cloning/\n",
             "%cd /content/ai-voice-cloning\n",
-            "!./setup-cuda.sh"
+            "# get local dependencies\n",
+            "!git submodule init\n",
+            "!git submodule update --remote\n",
+            "# setup venv\n",
+            "!python3 -m venv venv\n",
+            "!source ./venv/bin/activate\n",
+            "!python3 -m pip install --upgrade pip # just to be safe\n",
+            "# CUDA\n",
+            "!pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118\n",
+            "# install requirements\n",
+            "!python3 -m pip install -r ./modules/tortoise-tts/requirements.txt # install TorToiSe requirements\n",
+            "!python3 -m pip install -e ./modules/tortoise-tts/ # install TorToiSe\n",
+            "!python3 -m pip install -r ./modules/dlas/requirements.txt # instal DLAS requirements, last, because whisperx will break a dependency here\n",
+            "!python3 -m pip install -e ./modules/dlas/ # install DLAS\n",
+            "!python3 -m pip install -r ./requirements.txt # install local requirements"
          ]
       },
       {
@@ -115,7 +129,8 @@
          "cell_type":"code",
          "source":[
             "%cd /content/ai-voice-cloning/\n",
-            "!./start.sh --share"
+            "!source ./venv/bin/activate\n",
+            "!python3 ./src/main.py --share"
          ],
          "metadata":{
             "id":"QRA8jF3cF-YJ"

requirements.txt

@@ -1,5 +1,9 @@
-git+https://github.com/openai/whisper.git
+--extra-index-url https://download.pytorch.org/whl/cu118
+torch>=2.1.0
+torchvision
+torchaudio
 
+openai-whisper
 more-itertools
 ffmpeg-python
 gradio<=3.23.0

src/webui.py

@@ -92,7 +92,7 @@ def generate_proxy(
 			unload_tts()
 
 		raise e
-	
+
 	return (
 		outputs[0],
 		gr.update(value=sample, visible=sample is not None),
@@ -131,7 +131,7 @@ def history_view_results( voice ):
 		metadata, _ = read_generate_settings(f"{outdir}/{file}", read_latents=False)
 		if metadata is None:
 			continue
-			
+
 		values = []
 		for k in HISTORY_HEADERS:
 			v = file
@@ -185,7 +185,7 @@ def read_generate_settings_proxy(file, saveAs='.temp'):
 		os.makedirs(outdir, exist_ok=True)
 		with open(f'{outdir}/cond_latents.pth', 'wb') as f:
 			f.write(latents)
-		
+
 		latents = f'{outdir}/cond_latents.pth'
 
 	return (
@@ -229,7 +229,7 @@ def prepare_all_datasets( language, validation_text_length, validation_audio_len
 			print("Processing:", voice)
 			message = slice_dataset( voice, trim_silence=trim_silence, start_offset=slice_start_offset, end_offset=slice_end_offset, results=None, progress=progress )
 			messages.append(message)
-			
+
 	for voice in voices:
 		print("Processing:", voice)
 		message = prepare_dataset( voice, use_segments=slice_audio, text_length=validation_text_length, audio_length=validation_audio_length, progress=progress )
@@ -239,7 +239,7 @@ def prepare_all_datasets( language, validation_text_length, validation_audio_len
 
 def prepare_dataset_proxy( voice, language, validation_text_length, validation_audio_length, skip_existings, slice_audio, trim_silence, slice_start_offset, slice_end_offset, progress=gr.Progress(track_tqdm=True) ):
 	messages = []
-	
+
 	message = transcribe_dataset( voice=voice, language=language, skip_existings=skip_existings, progress=progress )
 	messages.append(message)
 
@@ -355,7 +355,7 @@ def setup_gradio():
 	voice_list_with_defaults = get_voice_list(append_defaults=True)
 	voice_list = get_voice_list()
 	result_voices = get_voice_list(args.results_folder)
-	
+
 	valle_models = get_valle_models()
 
 	autoregressive_models = get_autoregressive_models()
@@ -371,7 +371,9 @@ def setup_gradio():
 		arg = GENERATE_SETTINGS_ARGS[i]
 		GENERATE_SETTINGS[arg] = None
 
-	with gr.Blocks() as ui:
+	with gr.Blocks(theme="freddyaboulton/dracula_revamped", css="footer { display: none!important}", title="Voice Clonning WebUI") as ui:
+		gr.Markdown("## 🤗🎙️ Voice clonning ")
+		gr.Markdown("Ai Voice clonning <a href='https://git.ecker.tech/terminator/ai-voice-cloning-terminator'>based on Tortoise</a>")
 		with gr.Tab("Generate"):
 			with gr.Row():
 				with gr.Column():
@@ -402,7 +404,7 @@ def setup_gradio():
 						outputs=GENERATE_SETTINGS["mic_audio"],
 					)
 				with gr.Column():
-					preset = None						
+					preset = None
 					GENERATE_SETTINGS["candidates"] = gr.Slider(value=1, minimum=1, maximum=6, step=1, label="Candidates", visible=args.tts_backend=="tortoise")
 					GENERATE_SETTINGS["seed"] = gr.Number(value=0, precision=0, label="Seed", visible=args.tts_backend=="tortoise")
 
@@ -412,7 +414,7 @@ def setup_gradio():
 					GENERATE_SETTINGS["diffusion_iterations"] = gr.Slider(value=30, minimum=0, maximum=512, step=1, label="Iterations", visible=args.tts_backend=="tortoise")
 
 					GENERATE_SETTINGS["temperature"] = gr.Slider(value=0.95 if args.tts_backend=="vall-e" else 0.2, minimum=0, maximum=1, step=0.05, label="Temperature")
-					
+
 					show_experimental_settings = gr.Checkbox(label="Show Experimental Settings", visible=args.tts_backend=="tortoise")
 					reset_generate_settings_button = gr.Button(value="Reset to Default")
 				with gr.Column(visible=False) as col:
@@ -514,7 +516,7 @@ def setup_gradio():
 						transcribe_button = gr.Button(value="Transcribe and Process")
 						transcribe_all_button = gr.Button(value="Transcribe All")
 						diarize_button = gr.Button(value="Diarize", visible=False)
-						
+
 						with gr.Row():
 							slice_dataset_button = gr.Button(value="(Re)Slice Audio")
 							prepare_dataset_button = gr.Button(value="(Re)Create Dataset")
@@ -534,7 +536,7 @@ def setup_gradio():
 							TRAINING_SETTINGS["learning_rate"] = gr.Slider(label="Learning Rate", value=1e-5, minimum=0, maximum=1e-4, step=1e-6)
 							TRAINING_SETTINGS["mel_lr_weight"] = gr.Slider(label="Mel LR Ratio", value=1.00, minimum=0, maximum=1)
 							TRAINING_SETTINGS["text_lr_weight"] = gr.Slider(label="Text LR Ratio", value=0.01, minimum=0, maximum=1)
-							
+
 						with gr.Row(visible=args.tts_backend=="tortoise"):
 							lr_schemes = list(LEARNING_RATE_SCHEMES.keys())
 							TRAINING_SETTINGS["learning_rate_scheme"] = gr.Radio(lr_schemes, label="Learning Rate Scheme", value=lr_schemes[0], type="value")
@@ -567,7 +569,7 @@ def setup_gradio():
 
 						TRAINING_SETTINGS["source_model"] = gr.Dropdown( choices=autoregressive_models, label="Source Model", type="value", value=autoregressive_models[0], visible=args.tts_backend=="tortoise" )
 						TRAINING_SETTINGS["resume_state"] = gr.Textbox(label="Resume State Path", placeholder="./training/${voice}/finetune/training_state/${last_state}.state", visible=args.tts_backend=="tortoise")
-						
+
 						TRAINING_SETTINGS["voice"] = gr.Dropdown( choices=dataset_list, label="Dataset", type="value", value=dataset_list[0] if len(dataset_list) else ""  )
 
 						with gr.Row():
@@ -585,9 +587,9 @@ def setup_gradio():
 						refresh_configs = gr.Button(value="Refresh Configurations")
 						training_output = gr.TextArea(label="Console Output", interactive=False, max_lines=8)
 						verbose_training = gr.Checkbox(label="Verbose Console Output", value=True)
-						
+
 						keep_x_past_checkpoints = gr.Slider(label="Keep X Previous States", minimum=0, maximum=8, value=0, step=1)
-						
+
 						with gr.Row():
 							training_graph_x_min = gr.Number(label="X Min", precision=0, value=0)
 							training_graph_x_max = gr.Number(label="X Max", precision=0, value=0)
@@ -598,8 +600,8 @@ def setup_gradio():
 							start_training_button = gr.Button(value="Train")
 							stop_training_button = gr.Button(value="Stop")
 							reconnect_training_button = gr.Button(value="Reconnect")
-						
-						
+
+
 					with gr.Column():
 						training_loss_graph = gr.LinePlot(label="Training Metrics",
 							x="it", # x="epoch",
@@ -658,7 +660,7 @@ def setup_gradio():
 
 					EXEC_SETTINGS['results_folder'] = gr.Textbox(label="Results Folder", value=args.results_folder)
 					# EXEC_SETTINGS['tts_backend'] = gr.Dropdown(TTSES, label="TTS Backend", value=args.tts_backend if args.tts_backend else TTSES[0])
-					
+
 				if args.tts_backend=="vall-e":
 					with gr.Column():
 						EXEC_SETTINGS['valle_model'] = gr.Dropdown(choices=valle_models, label="VALL-E Model Config", value=args.valle_model if args.valle_model else valle_models[0])
@@ -668,7 +670,7 @@ def setup_gradio():
 					EXEC_SETTINGS['diffusion_model'] = gr.Dropdown(choices=diffusion_models, label="Diffusion Model", value=args.diffusion_model if args.diffusion_model else diffusion_models[0])
 					EXEC_SETTINGS['vocoder_model'] = gr.Dropdown(VOCODERS, label="Vocoder", value=args.vocoder_model if args.vocoder_model else VOCODERS[-1])
 					EXEC_SETTINGS['tokenizer_json'] = gr.Dropdown(tokenizer_jsons, label="Tokenizer JSON Path", value=args.tokenizer_json if args.tokenizer_json else tokenizer_jsons[0])
-					
+
 					EXEC_SETTINGS['training_default_halfp'] = TRAINING_SETTINGS['half_p']
 					EXEC_SETTINGS['training_default_bnb'] = TRAINING_SETTINGS['bitsandbytes']
 
@@ -718,7 +720,7 @@ def setup_gradio():
 				exec_inputs = list(EXEC_SETTINGS.values())
 				for k in EXEC_SETTINGS:
 					EXEC_SETTINGS[k].change( fn=update_args_proxy, inputs=exec_inputs )
-				
+
 				EXEC_SETTINGS['autoregressive_model'].change(
 					fn=update_autoregressive_model,
 					inputs=EXEC_SETTINGS['autoregressive_model'],
@@ -790,7 +792,7 @@ def setup_gradio():
 			],
 			outputs=GENERATE_SETTINGS['voice'],
 		)
-		
+
 		GENERATE_SETTINGS['emotion'].change(
 			fn=lambda value: gr.update(visible=value == "Custom"),
 			inputs=GENERATE_SETTINGS['emotion'],
@@ -944,7 +946,7 @@ def setup_gradio():
 			],
 			outputs=prepare_dataset_output
 		)
-		
+
 		training_refresh_dataset.click(
 			lambda: gr.update(choices=get_dataset_list()),
 			inputs=None,
@@ -966,7 +968,7 @@ def setup_gradio():
 
 		if os.path.isfile('./config/generate.json'):
 			ui.load(import_generate_settings_proxy, inputs=None, outputs=generate_settings)
-		
+
 		if args.check_for_updates:
 			ui.load(check_for_updates)