Merge pull request #1227 from IAHispano/fixes

Fixes

Author: Vidalnt

core.py

@@ -356,7 +356,9 @@ def run_tts_script(
             ],
         ),
     ]
-    subprocess.run(command_tts, check=True)
+    result = subprocess.run(command_tts, capture_output=True, text=True)
+    if result.returncode != 0:
+        raise RuntimeError(result.stderr.strip())
     infer_pipeline = import_voice_converter()
     infer_pipeline.convert_audio(
         pitch=pitch,
@@ -539,7 +541,11 @@ def run_train_script(
             ],
         ),
     ]
-    subprocess.run(command)
+    result = subprocess.run(command)
+    if result.returncode != 0:
+        raise RuntimeError(
+            f"Training failed for model {model_name}. Please check the console logs for more details."
+        )
     run_index_script(model_name, index_algorithm)
     return f"Model {model_name} trained successfully."
 

requirements.txt

@@ -3,25 +3,25 @@ setuptools==81.0.0
 pip>=23.3; sys_platform == 'darwin'
 wheel; sys_platform == 'darwin'
 PyYAML; sys_platform == 'darwin'
-numpy==2.3.5
-requests>=2.31.0,<2.32.0
+numpy==2.4.4
+requests
 tqdm
 wget
 
 # Audio processing
 ffmpeg-python>=0.2.0
 faiss-cpu==1.13.2
 librosa==0.11.0
-scipy==1.16.3
-soundfile==0.12.1
+scipy==1.17.1
+soundfile==0.13.1
 noisereduce
 pedalboard
 stftpitchshift
 soxr
 
 # Machine learning and deep learning
 omegaconf>=2.0.6; sys_platform == 'darwin'
-numba==0.63.1
+numba==0.65.0
 torch==2.7.1; sys_platform == 'darwin'
 torch==2.7.1+cu128; sys_platform == 'linux' or sys_platform == 'win32'
 torchaudio==2.7.1; sys_platform == 'darwin'
@@ -31,19 +31,20 @@ torchvision==0.22.1+cu128; sys_platform == 'linux' or sys_platform == 'win32'
 torchcrepe
 torchfcpe
 einops
-transformers==4.44.2
+transformers==5.4.0
 
 # Visualization and UI
 matplotlib==3.10.8
 tensorboard
-gradio==6.5.1
+gradio==6.10.0
 
 # Miscellaneous utilities
-certifi>=2023.07.22; sys_platform == 'darwin'
-antlr4-python3-runtime==4.8; sys_platform == 'darwin'
+certifi>=2026.2.25; sys_platform == 'darwin'
+antlr4-python3-runtime==4.13.2; sys_platform == 'darwin'
 tensorboardX
-edge-tts==7.2.7
+edge-tts==7.2.8
 pypresence
 beautifulsoup4
 sounddevice
 webrtcvad-wheels
+psutil

rvc/lib/tools/tts.py

@@ -1,6 +1,7 @@
 import asyncio
 import os
 import sys
+
 import edge_tts
 
 
@@ -21,10 +22,7 @@ async def main():
             with open(tts_file, "r") as file:
                 text = file.read()
 
-    try:
-        await edge_tts.Communicate(text, voice, rate=rates).save(output_file)
-    except Exception:
-        sys.exit(1)
+    await edge_tts.Communicate(text, voice, rate=rates).save(output_file)
 
 
 if __name__ == "__main__":

tabs/inference/inference.py

@@ -123,7 +123,6 @@ def get_files(type="model"):
 
 
 default_weight = next(iter(get_files("model")), None)
-index_files = sorted(get_files("index"))
 
 audio_paths = [
     os.path.join(root, name)
@@ -232,8 +231,6 @@ def change_choices(model):
 
     models_list = sorted(get_files("model"))
     indexes_list = sorted(get_files("index"))
-    if indexes_list:
-        indexes_list = [("No Index", "")] + [(x, x) for x in indexes_list]
 
     audio_paths = [
         os.path.join(root, name)
@@ -487,8 +484,6 @@ def filter_dropdowns(filter_text):
     all_indexes = sorted(get_files("index"))
     filtered_models = [m for m in all_models if ft in m.lower()]
     filtered_indexes = [i for i in all_indexes if ft in i.lower()]
-    if filtered_indexes:
-        filtered_indexes = [("No Index", "")] + [(x, x) for x in filtered_indexes]
     return (gr.update(choices=filtered_models), gr.update(choices=filtered_indexes))
 
 
@@ -525,11 +520,7 @@ def inference_tab():
             index_file = gr.Dropdown(
                 label=i18n("Index File"),
                 info=i18n("Select the index file to use for the conversion."),
-                choices=(
-                    [("No Index", "")] + [(x, x) for x in index_files]
-                    if index_files
-                    else []
-                ),
+                choices=sorted(get_files("index")),
                 value=match_index(default_weight),
                 interactive=True,
                 allow_custom_value=True,

tabs/realtime/realtime.py

@@ -1181,11 +1181,7 @@ def realtime_tab():
                     index_choices = sorted(get_files("index"))
                     index_file = gr.Dropdown(
                         label=i18n("Index File"),
-                        choices=(
-                            [("No Index", "")] + [(x, x) for x in index_choices]
-                            if index_choices
-                            else []
-                        ),
+                        choices=index_choices,
                         value=get_safe_index_value(
                             saved_settings["index_file"],
                             index_choices,
@@ -2205,8 +2201,6 @@ def delay_visible(checkbox):
             def refresh_all():
                 new_names = get_files("model")
                 new_indexes = sorted(get_files("index"))
-                if new_indexes:
-                    new_indexes = [("No Index", "")] + [(x, x) for x in new_indexes]
                 return (
                     gr.update(choices=sorted(new_names, key=extract_model_and_epoch)),
                     gr.update(choices=new_indexes),
@@ -2224,8 +2218,6 @@ def refresh_all():
             def refresh_all():
                 new_names = get_files("model")
                 new_indexes = sorted(get_files("index"))
-                if new_indexes:
-                    new_indexes = [("No Index", "")] + [(x, x) for x in new_indexes]
                 input_choices, output_choices = get_audio_devices_formatted()
                 input_choices, output_choices = list(input_choices.keys()), list(
                     output_choices.keys()

tabs/settings/sections/restart.py

@@ -1,22 +1,50 @@
-import gradio as gr
+import json
 import os
 import sys
-import json
+
+import gradio as gr
+import psutil
 
 now_dir = os.getcwd()
 
 
 def stop_train(model_name: str):
+    if not model_name or model_name == "":
+        return
+
     pid_file_path = os.path.join(now_dir, "logs", model_name, "config.json")
+    killed = 0
+
     try:
         with open(pid_file_path, "r", encoding="utf-8") as pid_file:
             pid_data = json.load(pid_file)
             pids = pid_data.get("process_pids", [])
-        with open(pid_file_path, "w") as pid_file:
+
+        with open(pid_file_path, "w", encoding="utf-8") as pid_file:
             pid_data.pop("process_pids", None)
             json.dump(pid_data, pid_file, indent=4)
+
         for pid in pids:
-            os.kill(pid, 9)
+            try:
+                parent = psutil.Process(pid)
+                for child in parent.children(recursive=True):
+                    child.kill()
+                parent.kill()
+                killed += 1
+            except psutil.NoSuchProcess:
+                pass
+            except Exception:
+                try:
+                    os.kill(pid, 9)
+                    killed += 1
+                except:
+                    pass
+
+        # if killed > 0:
+        #    gr.Info(f"Training stopped successfully ({killed} process(es) terminated)")
+        # else:
+        #    gr.Info("No active training processes found")
+
     except:
         pass
 
@@ -25,9 +53,22 @@ def stop_infer():
     pid_file_path = os.path.join(now_dir, "assets", "infer_pid.txt")
     try:
         with open(pid_file_path, "r") as pid_file:
-            pids = [int(pid) for pid in pid_file.readlines()]
+            pids = [int(pid) for pid in pid_file.readlines() if pid.strip()]
+
         for pid in pids:
-            os.kill(pid, 9)
+            try:
+                parent = psutil.Process(pid)
+                for child in parent.children(recursive=True):
+                    child.kill()
+                parent.kill()
+            except psutil.NoSuchProcess:
+                pass
+            except Exception:
+                try:
+                    os.kill(pid, 9)
+                except:
+                    pass
+
         os.remove(pid_file_path)
     except:
         pass

tabs/train/train.py

@@ -768,7 +768,10 @@ def enforce_terms(terms_accepted, *args):
                 message = "You must agree to the Terms of Use to proceed."
                 gr.Info(message)
                 return message
-            return run_train_script(*args)
+            try:
+                return run_train_script(*args)
+            except Exception as e:
+                return e
 
         terms_checkbox = gr.Checkbox(
             label=i18n("I agree to the terms of use"),
@@ -788,32 +791,6 @@ def enforce_terms(terms_accepted, *args):
 
         with gr.Row():
             train_button = gr.Button(i18n("Start Training"))
-            train_button.click(
-                fn=enforce_terms,
-                inputs=[
-                    terms_checkbox,
-                    model_name,
-                    save_every_epoch,
-                    save_only_latest,
-                    save_every_weights,
-                    total_epoch,
-                    sampling_rate,
-                    batch_size,
-                    gpu,
-                    overtraining_detector,
-                    overtraining_threshold,
-                    pretrained,
-                    cleanup,
-                    index_algorithm,
-                    cache_dataset_in_gpu,
-                    custom_pretrained,
-                    g_pretrained_path,
-                    d_pretrained_path,
-                    vocoder,
-                    checkpointing,
-                ],
-                outputs=[train_output_info],
-            )
 
             stop_train_button = gr.Button(i18n("Stop Training"), visible=False)
             stop_train_button.click(
@@ -880,31 +857,21 @@ def enforce_terms(terms_accepted, *args):
                     )
 
             def toggle_visible(checkbox):
-                return {"visible": checkbox, "__type__": "update"}
+                return gr.update(visible=checkbox)
 
             def toggle_pretrained(pretrained, custom_pretrained):
-                if custom_pretrained == False:
-                    return {"visible": pretrained, "__type__": "update"}, {
-                        "visible": False,
-                        "__type__": "update",
-                    }
+                if not custom_pretrained:
+                    return gr.update(visible=pretrained), gr.update(visible=False)
                 else:
-                    return {"visible": pretrained, "__type__": "update"}, {
-                        "visible": pretrained,
-                        "__type__": "update",
-                    }
+                    return gr.update(visible=pretrained), gr.update(visible=pretrained)
 
-            def enable_stop_train_button():
-                return {"visible": False, "__type__": "update"}, {
-                    "visible": True,
-                    "__type__": "update",
-                }
+            def enable_stop_train_button(terms_accepted):
+                if not terms_accepted:
+                    return gr.update(visible=True), gr.update(visible=False)
+                return gr.update(visible=False), gr.update(visible=True)
 
             def disable_stop_train_button():
-                return {"visible": True, "__type__": "update"}, {
-                    "visible": False,
-                    "__type__": "update",
-                }
+                return gr.update(visible=True), gr.update(visible=False)
 
             def download_prerequisites():
                 gr.Info(
@@ -1030,11 +997,38 @@ def update_slider_visibility(noise_reduction):
                 inputs=[overtraining_detector],
                 outputs=[overtraining_settings],
             )
+
             train_button.click(
                 fn=enable_stop_train_button,
-                inputs=[],
+                inputs=[terms_checkbox],
                 outputs=[train_button, stop_train_button],
+            ).then(
+                fn=enforce_terms,
+                inputs=[
+                    terms_checkbox,
+                    model_name,
+                    save_every_epoch,
+                    save_only_latest,
+                    save_every_weights,
+                    total_epoch,
+                    sampling_rate,
+                    batch_size,
+                    gpu,
+                    overtraining_detector,
+                    overtraining_threshold,
+                    pretrained,
+                    cleanup,
+                    index_algorithm,
+                    cache_dataset_in_gpu,
+                    custom_pretrained,
+                    g_pretrained_path,
+                    d_pretrained_path,
+                    vocoder,
+                    checkpointing,
+                ],
+                outputs=[train_output_info],
             )
+
             train_output_info.change(
                 fn=disable_stop_train_button,
                 inputs=[],

tabs/tts/tts.py

@@ -19,7 +19,6 @@
     filter_dropdowns,
     get_files,
     get_speakers_id,
-    index_files,
     match_index,
     refresh_embedders_folders,
     update_filter_visibility,
@@ -73,11 +72,7 @@ def tts_tab():
             index_file = gr.Dropdown(
                 label=i18n("Index File"),
                 info=i18n("Select the index file to use for the conversion."),
-                choices=(
-                    [("No Index", "")] + [(x, x) for x in index_files]
-                    if index_files
-                    else []
-                ),
+                choices=sorted(get_files("index")),
                 value=match_index(default_weight),
                 interactive=True,
                 allow_custom_value=True,