fix: resolve 24 bugs across codebase

BF667 · BF667 · commit 527526446f55 · 2026-05-18T22:13:55.000Z
CRITICAL fixes (4):
- easy_gui.py: Remove misplaced module-level code inside function body (IndentationError)
- gui.py: Move argparse/sys.exit inside __name__ == '__main__' guard
- cli.py: Remove misplaced module-level code inside try block (IndentationError)
- cli_complete.py: Remove call to undefined main(), fix --help_create_reference prefix

HIGH fixes (11):
- model_utils.py: Fix fusion crash (.shape on dict), fix ONNX model_info crash (None.get)
- csrt.py: Fix milliseconds always computed as 0
- realtime_client.py: Guard FastAPI() instantiation when package not installed
- pipeline.py: Fix None slice in audio_pad (inserts axis instead of full array)
- audio_processing.py: Fix STFT cache destroyed on every other call
- create_reference.py: Fix shutil.rmtree deleting wrong directory
- realtime/pipeline.py: Fix TypeError when return_length is None
- realtime/audio.py: Fix output ASIO settings checking input device instead of output
- inference.py: Fix NameError in finally (unbound segments), fix unbound end_time

MEDIUM fixes (8):
- config.py: Fix is_fp16() corrupting config.json with absolute paths
- configs/__init__.py: Remove undefined config_settings from __all__
- config.json: Fix typo fushion_tab -&gt; fusion_tab
- pyproject.toml: Fix Python version range mismatch with _version.py
- feedback.py: Add .onnx to model deletion filter
- gdown.py: Fix AttributeError on unexpected Content-Disposition, fix file handle leak
- huggingface.py: Fix PEP 8 None comparison
- create_index.py: Fix n_ivf=0 producing invalid FAISS index
diff --git a/advanced_rvc_inference/__init__.py b/advanced_rvc_inference/__init__.py
diff --git a/advanced_rvc_inference/api/cli.py b/advanced_rvc_inference/api/cli.py
@@ -682,11 +682,6 @@ def cmd_serve(args):
     try:
         from advanced_rvc_inference.app.gui import launch
 
-# Fix import errors by ensuring cwd is in sys.path
-import os
-if os.getcwd() not in sys.path:
-    sys.path.append(os.getcwd())
-
         launch(
             share=args.share,
             server_name=args.host,
diff --git a/advanced_rvc_inference/api/cli_complete.py b/advanced_rvc_inference/api/cli_complete.py
@@ -26,7 +26,7 @@
     "--help_audio_effects", "--help_convert", "--help_create_dataset",
     "--help_create_index", "--help_extract", "--help_preprocess",
     "--help_separate_music", "--help_train", "--help",
-    "--create_reference", "help_create_reference",
+    "--create_reference", "--help_create_reference",
 ]
 
 if argv not in argv_is_allows:
@@ -331,5 +331,3 @@
         mp.set_start_method("spawn")
     if "--preprocess" in argv or "--extract" in argv:
         mp.set_start_method("spawn", force=True)
-
-    main()
diff --git a/advanced_rvc_inference/app/easy_gui.py b/advanced_rvc_inference/app/easy_gui.py
@@ -655,10 +655,6 @@ def easy_download(url, name):
                 def get_gpu_info():
                     import torch
 
-# Fix import errors by ensuring cwd is in sys.path
-import os
-if os.getcwd() not in sys.path:
-    sys.path.append(os.getcwd())
                     if torch.cuda.is_available():
                         name = torch.cuda.get_device_name(0)
                         mem = torch.cuda.get_device_properties(0).total_memory / (1024**3)
diff --git a/advanced_rvc_inference/app/gui.py b/advanced_rvc_inference/app/gui.py
@@ -254,7 +254,7 @@ def get_package_assets_path():
         return 1
     except OSError as e:
         if "Address already in use" in str(e):
-            logger.error(f"Port {server_port} is already in use")
+            logger.error(f"Port {port} is already in use")
             logger.info("Try using a different port with: rvc-cli serve --port 7861")
         else:
             logger.error(f"Failed to start server: {e}")
@@ -335,19 +335,14 @@ def create_app():
 if __name__ == "__main__":
     import argparse
 
-# Fix import errors by ensuring cwd is in sys.path
-import os
-if os.getcwd() not in sys.path:
-    sys.path.append(os.getcwd())
-
     parser = argparse.ArgumentParser(description="Launch Advanced RVC Inference GUI")
     parser.add_argument("--host", default="0.0.0.0", help="Host to bind to")
     parser.add_argument("--port", type=int, default=7860, help="Port to bind to")
     parser.add_argument("--share", action="store_true", help="Create public URL (may fail in some environments)")
     parser.add_argument("--no-share", action="store_true", help="Disable public URL, use local access only")
     parser.add_argument("--open", action="store_true", help="Open in browser")
     parser.add_argument("--keep-alive", action="store_true", default=True, help="Keep tunnel alive (default: True)")
-    parser.add_argument("--easy", "-ez", type=str, default=None, help="Launch Easy GUI (simplified mode). Use 'true' to enable.")
+    parser.add_argument("--easy", "-E", type=str, default=None, help="Launch Easy GUI (simplified mode). Use 'true' to enable.")
 
     args = parser.parse_args()
     easy_mode = args.easy is not None and args.easy.lower() in ("true", "1", "yes")
diff --git a/advanced_rvc_inference/app/tabs/extra/extra.py b/advanced_rvc_inference/app/tabs/extra/extra.py
@@ -18,7 +18,7 @@ def extra_tab(app):
             gr.Markdown(translations["settings_markdown"])
             settings_tab(app)
             
-        with gr.TabItem(translations["fushion"], visible=configs.get("fushion_tab", True)):
+        with gr.TabItem(translations["fushion"], visible=configs.get("fusion_tab", True)):
             gr.Markdown(translations["fushion_markdown"])
             fushion_tab()
 
diff --git a/advanced_rvc_inference/configs/__init__.py b/advanced_rvc_inference/configs/__init__.py
@@ -4,7 +4,7 @@
 This package contains configuration files and settings.
 """
 
-__all__ = ["config", "config_settings"]
+__all__ = ["config"]
 
 from .config import Config
 from . import config as config_module
diff --git a/advanced_rvc_inference/configs/config.json b/advanced_rvc_inference/configs/config.json
@@ -561,7 +561,7 @@
     "tts_tab": true,
     "create_dataset_tab": true,
     "training_tab": true,
-    "fushion_tab": true,
+    "fusion_tab": true,
     "read_tab": true,
     "onnx_tab": true,
     "downloads_tab": true,
diff --git a/advanced_rvc_inference/configs/config.py b/advanced_rvc_inference/configs/config.py
@@ -159,7 +159,7 @@ def is_fp16(self):
             fp16 = False
             try:
                 with open(self.configs_path, "w", encoding="utf-8") as f:
-                    json.dump(self.configs, f, indent=4)
+                    json.dump({"fp16": False}, f)
             except OSError:
                 pass
 
diff --git a/advanced_rvc_inference/engine/inference/audio_processing.py b/advanced_rvc_inference/engine/inference/audio_processing.py
@@ -164,8 +164,9 @@ def spectral_subtract_denoise(audio, sr, noise_seconds=0.4, alpha=1.0, n_fft=102
 
     if stft is None and device.startswith(("ocl", "privateuseone")):
         from advanced_rvc_inference.engine.models.backends.utils import STFT
-        stft = STFT(filter_length=n_fft, hop_length=hop_length, win_length=None, window="hann").to(device) 
-    else: stft = None
+        stft = STFT(filter_length=n_fft, hop_length=hop_length, win_length=None, window="hann").to(device)
+    elif not device.startswith(("ocl", "privateuseone")):
+        stft = None
 
     x = torch.from_numpy(audio.astype(np.float32)).float().unsqueeze(0).to(device)
     window = torch.hann_window(n_fft).to(device)
diff --git a/advanced_rvc_inference/engine/inference/create_reference.py b/advanced_rvc_inference/engine/inference/create_reference.py
@@ -90,7 +90,7 @@ def create_reference(
         sys.exit(1)
 
     output_reference = os.path.join(configs["reference_path"], f"{reference_name}_{version}_{embedder_model}_{pitch_guidance}_{use_energy}")
-    if os.path.exists(output_reference): shutil.rmtree(reference_name, ignore_errors=True)
+    if os.path.exists(output_reference): shutil.rmtree(output_reference, ignore_errors=True)
 
     os.makedirs(output_reference)
     logger.info(translations["start_create_reference"])
diff --git a/advanced_rvc_inference/engine/inference/inference.py b/advanced_rvc_inference/engine/inference/inference.py
@@ -745,7 +745,10 @@ def whisper_process(
     except Exception as e:
         out_queue.put(e)
     finally:
-        del segments
+        try:
+            del segments
+        except NameError:
+            pass
         gc.collect()
 
 def convert_with_whisper(
@@ -925,6 +928,7 @@ def extract_number(filename):
 
         merged_segments, current_text = [], []
         current_speaker, current_start = None, None
+        end_time = 0.0
 
         for i, segment in enumerate(segments):
             speaker = segment["speaker"]
diff --git a/advanced_rvc_inference/engine/inference/pipeline.py b/advanced_rvc_inference/engine/inference/pipeline.py
@@ -182,7 +182,7 @@ def pipeline(self, logger, model, net_g, sid, audio, f0_up_key, f0_method, file_
                 model, 
                 net_g, 
                 sid, 
-                audio_pad[t:], 
+                audio_pad[t if t is not None else 0:], 
                 (pitch[:, t // self.window :] if t is not None else pitch) if pitch_guidance else None, 
                 (pitchf[:, t // self.window :] if t is not None else pitchf) if pitch_guidance else None, 
                 index, 
diff --git a/advanced_rvc_inference/engine/realtime/audio.py b/advanced_rvc_inference/engine/realtime/audio.py
@@ -237,7 +237,7 @@ def start(self, input_device_id, output_device_id, output_monitor_id, exclusive_
 
         if output_audio_device and "WASAPI" in output_audio_device.host_api:
             output_extra_setting = sd.WasapiSettings(exclusive=wasapi_exclusive_mode, auto_convert=not wasapi_exclusive_mode)
-        elif input_audio_device and "ASIO" in input_audio_device.host_api and asio_output_channel != -1:
+        elif output_audio_device and "ASIO" in output_audio_device.host_api and asio_output_channel != -1:
             output_extra_setting = sd.AsioSettings(channel_selectors=[asio_output_channel])
             output_channels = 1
 
diff --git a/advanced_rvc_inference/engine/realtime/pipeline.py b/advanced_rvc_inference/engine/realtime/pipeline.py
@@ -90,6 +90,8 @@ def __init__(self, inference, embedder, predictor = None, rms = None, index = (N
     def execute(self, audio, pitch = None, pitchf = None, f0_up_key = 0, index_rate = 0.5, audio_feats_len = 0, silence_front = 0, skip_head = None, return_length = None, protect = 0.5, filter_radius = 3, rms_mix_rate = 1, f0_autotune = False, f0_autotune_strength = 1, proposal_pitch = False, proposal_pitch_threshold = 255.0):
         with torch.no_grad():     
             assert audio.dim() == 1, audio.dim()
+            if return_length is None or return_length == 0:
+                return_length = audio_feats_len if audio_feats_len > 0 else 1
             formant_length = int(np.ceil(return_length * 1.0))
 
             pitch, pitchf = self.predictor.realtime_calculator(audio[silence_front:], self.f0_method, pitch, pitchf, f0_up_key, filter_radius, f0_autotune, f0_autotune_strength, proposal_pitch, proposal_pitch_threshold) if self.use_f0 else (None, None)
diff --git a/advanced_rvc_inference/engine/training/create_index.py b/advanced_rvc_inference/engine/training/create_index.py
@@ -46,7 +46,7 @@ def main():
         if big_npy.shape[0] > 2e5 and (index_algorithm == "Auto" or index_algorithm == "KMeans"): big_npy = (MiniBatchKMeans(n_clusters=10000, verbose=True, batch_size=256 * cpu_count(), compute_labels=False, init="random").fit(big_npy).cluster_centers_)
         np.save(os.path.join(exp_dir, "total_fea.npy"), big_npy)
 
-        n_ivf = min(int(16 * np.sqrt(big_npy.shape[0])), big_npy.shape[0] // 39)
+        n_ivf = max(1, min(int(16 * np.sqrt(big_npy.shape[0])), big_npy.shape[0] // 39))
         index_trained = faiss.index_factory(256 if version == "v1" else 768, f"IVF{n_ivf},Flat")
         index_ivf_trained = faiss.extract_index_ivf(index_trained)
         index_ivf_trained.nprobe = 1
diff --git a/advanced_rvc_inference/services/csrt.py b/advanced_rvc_inference/services/csrt.py
@@ -68,7 +68,8 @@ def format_timestamp(seconds):
     hours = int(seconds // 3600)
     minutes = int((seconds % 3600) // 60)
 
-    seconds = int(seconds % 60)
-    miliseconds = int((seconds - int(seconds)) * 1000)
+    fractional = seconds % 60
+    seconds = int(fractional)
+    milliseconds = int((fractional - seconds) * 1000)
 
-    return f"{hours:02}:{minutes:02}:{seconds:02},{miliseconds:03}"
+    return f"{hours:02}:{minutes:02}:{seconds:02},{milliseconds:03}"
diff --git a/advanced_rvc_inference/services/model_utils.py b/advanced_rvc_inference/services/model_utils.py
@@ -50,8 +50,8 @@ def extract(ckpt):
         vocoder = ckpt1.get("vocoder", "Default")
         rms_extract = ckpt1.get("energy", False)
 
-        ckpt1 = extract(ckpt1) if "model" in ckpt1 else ckpt1["weight"]
-        ckpt2 = extract(ckpt2) if "model" in ckpt2 else ckpt2["weight"]
+        ckpt1 = extract(ckpt1)["weight"] if "model" in ckpt1 else ckpt1["weight"]
+        ckpt2 = extract(ckpt2)["weight"] if "model" in ckpt2 else ckpt2["weight"]
 
         if sorted(list(ckpt1.keys())) != sorted(list(ckpt2.keys())): 
             gr_warning(translations["architectures_not_same"])
@@ -136,6 +136,9 @@ def prettify_date(date_str):
                 model_data = json.loads(prop.value)
                 break
 
+        if model_data is None:
+            return gr_warning(translations.get("model_info_not_found", "No model info found in ONNX metadata"))
+
     gr_info(translations["read_info"])
 
     epochs = model_data.get("epoch", None)
diff --git a/advanced_rvc_inference/services/realtime_client.py b/advanced_rvc_inference/services/realtime_client.py
diff --git a/advanced_rvc_inference/ui/feedback.py b/advanced_rvc_inference/ui/feedback.py
diff --git a/advanced_rvc_inference/utils/gdown.py b/advanced_rvc_inference/utils/gdown.py
diff --git a/advanced_rvc_inference/utils/huggingface.py b/advanced_rvc_inference/utils/huggingface.py
diff --git a/pyproject.toml b/pyproject.toml
