Uses tabs instead of columns

TTS/demos/xtts_ft_demo/requirements.txt

@@ -0,0 +1 @@
+faster_whisper

TTS/demos/xtts_ft_demo/xtts_demo.py

@@ -95,11 +95,8 @@ def read_logs():
 
 
 with gr.Blocks() as demo:
-    with gr.Tab("XTTS"):
+    state_vars = gr.State()
-        state_vars = gr.State(
+    with gr.Tab("Data processing"):    
-        )
-        with gr.Row():
-            with gr.Column() as col1:       
         upload_file = gr.Audio(
             sources="upload",
             label="Select here the audio files that you want to use for XTTS trainining !",
@@ -138,7 +135,36 @@ with gr.Blocks() as demo:
 
         prompt_compute_btn = gr.Button(value="Step 1 - Create dataset.")
     
-            with gr.Column() as col2:
+        def preprocess_dataset(audio_path, language, state_vars, progress=gr.Progress(track_tqdm=True)):
+            # create a temp directory to save the dataset
+            out_path = tempfile.TemporaryDirectory().name
+            if audio_path is None:
+                # ToDo: raise an error
+                pass
+            else:
+                train_meta, eval_meta = format_audio_list([audio_path], target_language=language, out_path=out_path, gradio_progress=progress)
+
+            state_vars = {}
+            state_vars["train_csv"] = train_meta
+            state_vars["eval_csv"] = eval_meta
+            print(state_vars)
+            return "Dataset Processed!", state_vars
+
+        prompt_compute_btn.click(
+            fn=preprocess_dataset,
+            inputs=[
+                upload_file,
+                lang,
+                state_vars,
+            ],
+            outputs=[
+                progress_data,
+                state_vars,
+            ],
+        )
+        
+
+    with gr.Tab("Fine-tuning XTTS"):
         num_epochs =  gr.Slider(
             label="num_epochs",
             minimum=1,
@@ -163,7 +189,24 @@ with gr.Blocks() as demo:
         demo.load(read_logs, None, logs_tts_train, every=1)
         train_btn = gr.Button(value="Step 2 - Run the training")
 
-            with gr.Column() as col3:
+        def train_model(language, num_epochs, batch_size, state_vars, output_path="./", progress=gr.Progress(track_tqdm=True)):
+            # state_vars = {'train_csv': '/tmp/tmprh4k_vou/metadata_train.csv', 'eval_csv': '/tmp/tmprh4k_vou/metadata_eval.csv'}
+
+            config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, state_vars["train_csv"], state_vars["eval_csv"], output_path=output_path)
+            # copy original files to avoid parameters changes issues
+            os.system(f"cp {config_path} {exp_path}")
+            os.system(f"cp {vocab_file} {exp_path}")
+
+            ft_xtts_checkpoint = os.path.join(exp_path, "best_model.pth")
+            state_vars["config_path"] = config_path
+            state_vars["original_xtts_checkpoint"] = original_xtts_checkpoint
+            state_vars["vocab_file"] = vocab_file
+            state_vars["ft_xtts_checkpoint"] = ft_xtts_checkpoint
+            state_vars["speaker_audio_file"] = speaker_wav
+            return "Model training done!", state_vars, config_path, vocab_file, ft_xtts_checkpoint, speaker_wav
+
+
+    with gr.Tab("Inference"):
         xtts_checkpoint = gr.Textbox(
             label="XTTS checkpoint path:",
             value="",
@@ -212,51 +255,6 @@ with gr.Blocks() as demo:
         reference_audio = gr.Audio(label="Reference audio used.")
 
 
-        def preprocess_dataset(audio_path, language, state_vars, progress=gr.Progress(track_tqdm=True)):
-            # create a temp directory to save the dataset
-            out_path = tempfile.TemporaryDirectory().name
-            if audio_path is None:
-                # ToDo: raise an error
-                pass
-            else:
-                train_meta, eval_meta = format_audio_list([audio_path], target_language=language, out_path=out_path, gradio_progress=progress)
-
-            state_vars = {}
-            state_vars["train_csv"] = train_meta
-            state_vars["eval_csv"] = eval_meta
-            print(state_vars)
-            return "Dataset Processed!", state_vars
-
-        prompt_compute_btn.click(
-            fn=preprocess_dataset,
-            inputs=[
-                upload_file,
-                lang,
-                state_vars,
-            ],
-            outputs=[
-                progress_data,
-                state_vars,
-            ],
-        )
-        
-        def train_model(language, num_epochs, batch_size, state_vars, output_path="./", progress=gr.Progress(track_tqdm=True)):
-            # state_vars = {'train_csv': '/tmp/tmprh4k_vou/metadata_train.csv', 'eval_csv': '/tmp/tmprh4k_vou/metadata_eval.csv'}
-
-            config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, state_vars["train_csv"], state_vars["eval_csv"], output_path=output_path)
-            # copy original files to avoid parameters changes issues
-            os.system(f"cp {config_path} {exp_path}")
-            os.system(f"cp {vocab_file} {exp_path}")
-
-            ft_xtts_checkpoint = os.path.join(exp_path, "best_model.pth")
-            state_vars["config_path"] = config_path
-            state_vars["original_xtts_checkpoint"] = original_xtts_checkpoint
-            state_vars["vocab_file"] = vocab_file
-            state_vars["ft_xtts_checkpoint"] = ft_xtts_checkpoint
-            state_vars["speaker_audio_file"] = speaker_wav
-            return "Model training done!", state_vars, config_path, vocab_file, ft_xtts_checkpoint, speaker_wav
-
-
         train_btn.click(
             fn=train_model,
             inputs=[
@@ -268,6 +266,7 @@ with gr.Blocks() as demo:
             outputs=[progress_train, state_vars, xtts_config, xtts_vocab, xtts_checkpoint, speaker_reference_audio],
         )
         
+
         tts_btn.click(
             fn=run_tts,
             inputs=[