chore: run autoflake and black formatting

Author: semnil

app.py

@@ -27,12 +27,15 @@
 # connection during asyncio shutdown (WinError 10054 / ProactorBasePipeTransport).
 if sys.platform == "win32":
     import asyncio.proactor_events as _pe
+
     _orig_ccl = _pe._ProactorBasePipeTransport._call_connection_lost
+
     def _ccl_patched(self, exc):
         try:
             _orig_ccl(self, exc)
         except ConnectionResetError:
             pass
+
     _pe._ProactorBasePipeTransport._call_connection_lost = _ccl_patched
 
 # Zluda hijack

rvc/realtime/audio.py

@@ -390,9 +390,7 @@ def start(
                 if asio_output_stereo
                 else [asio_output_channel]
             )
-            output_extra_setting = sd.AsioSettings(
-                channel_selectors=output_selectors
-            )
+            output_extra_setting = sd.AsioSettings(channel_selectors=output_selectors)
             output_channels = len(output_selectors)
 
         if self.use_monitor:
@@ -414,7 +412,9 @@ def start(
                 )
                 monitor_channels = 1
 
-        block_frame = int((read_chunk_size * 128 / AUDIO_SAMPLE_RATE) * audio_sample_rate)
+        block_frame = int(
+            (read_chunk_size * 128 / AUDIO_SAMPLE_RATE) * audio_sample_rate
+        )
 
         try:
             self.run_audio_stream(

rvc/realtime/callbacks.py

@@ -130,7 +130,10 @@ def change_voice(
             return audio, vol, [0, perf_ms, 0], None
 
         # No result ready yet; replay previous output to avoid underrun.
-        if self._last_output is not None and self._last_output.shape[0] == received_data.shape[0]:
+        if (
+            self._last_output is not None
+            and self._last_output.shape[0] == received_data.shape[0]
+        ):
             return self._last_output, self._last_vol, [0, 0, 0], None
 
         return np.zeros(received_data.shape[0], dtype=np.float32), 0, [0, 0, 0], None

rvc/realtime/core.py

@@ -366,9 +366,7 @@ def inference(
         )
 
         # Scale output by the current input RMS to suppress residue during silence.
-        audio_out: torch.Tensor = self.resample_out(
-            audio_model * vol_t
-        )
+        audio_out: torch.Tensor = self.resample_out(audio_model * vol_t)
         return audio_out, vol, False
 
     def __del__(self):
@@ -531,11 +529,7 @@ def process_audio(
             n_hops = block_size // hop
             if n_hops >= 1:
                 hop_energy = (
-                    audio[: n_hops * hop]
-                    .reshape(n_hops, hop)
-                    .abs()
-                    .max(dim=1)
-                    .values
+                    audio[: n_hops * hop].reshape(n_hops, hop).abs().max(dim=1).values
                 )
                 peak = hop_energy.max().item()
                 onset_sample = 0
@@ -562,7 +556,9 @@ def process_audio(
         # Pad if audio is shorter than block_size + crossfade_frame.
         _need = block_size + self.crossfade_frame
         if audio.shape[0] < _need:
-            pad = torch.zeros(_need - audio.shape[0], device=audio.device, dtype=audio.dtype)
+            pad = torch.zeros(
+                _need - audio.shape[0], device=audio.device, dtype=audio.dtype
+            )
             audio = torch.cat([audio, pad])
         self.sola_buffer[:] = audio[block_size : block_size + self.crossfade_frame]
         audio_output = audio[:block_size].detach().cpu().numpy()

rvc/realtime/pipeline.py

@@ -255,7 +255,11 @@ def voice_conversion(
             if self.use_f0:
                 # Extract F0 from the most recent audio window only.
                 shift = (block_size_16k or skip_head * self.window) // self.window
-                f0_frame = block_size_16k + 800 if block_size_16k else skip_head * self.window + 800
+                f0_frame = (
+                    block_size_16k + 800
+                    if block_size_16k
+                    else skip_head * self.window + 800
+                )
                 if self.f0_method == "rmvpe":
                     f0_frame = 5120 * ((f0_frame - 1) // 5120 + 1) - 160
                 f0_frame = min(f0_frame, audio.shape[0])
@@ -272,15 +276,17 @@ def voice_conversion(
                 )
                 # Remove batch dimension.
                 f0_coarse_new = f0_coarse_new.squeeze(0)
-                f0_new        = f0_new.squeeze(0)
+                f0_new = f0_new.squeeze(0)
 
                 # Shift pitch cache left by one block and append new frames (trimmed [3:-1]).
                 pitch[:-shift] = pitch[shift:].clone()
                 pitchf[:-shift] = pitchf[shift:].clone()
-                interior_coarse = f0_coarse_new[3:-1] if f0_coarse_new.shape[0] > 4 else f0_coarse_new
-                interior_f      = f0_new[3:-1]        if f0_new.shape[0] > 4        else f0_new
-                pitch[-interior_coarse.shape[0]:]  = interior_coarse
-                pitchf[-interior_f.shape[0]:]      = interior_f
+                interior_coarse = (
+                    f0_coarse_new[3:-1] if f0_coarse_new.shape[0] > 4 else f0_coarse_new
+                )
+                interior_f = f0_new[3:-1] if f0_new.shape[0] > 4 else f0_new
+                pitch[-interior_coarse.shape[0] :] = interior_coarse
+                pitchf[-interior_f.shape[0] :] = interior_f
             else:
                 pitch, pitchf = None, None
 
@@ -316,8 +322,10 @@ def voice_conversion(
                 feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(
                     0, 2, 1
                 )[:, :p_len, :]
-                pitch_p  = pitch[-p_len:].unsqueeze(0)
-                pitchf_p = pitchf[-p_len:].unsqueeze(0) * (formant_length / return_length)
+                pitch_p = pitch[-p_len:].unsqueeze(0)
+                pitchf_p = pitchf[-p_len:].unsqueeze(0) * (
+                    formant_length / return_length
+                )
 
                 # Pitch protection blending
                 if protect < 0.5:
@@ -342,7 +350,7 @@ def voice_conversion(
             ).float()
             # Match output RMS to the current block's input RMS.
             if volume_envelope < 1:
-                rms_src = audio[-(return_length * self.window):].cpu().numpy()
+                rms_src = audio[-(return_length * self.window) :].cpu().numpy()
                 out_audio = AudioProcessor.change_rms(
                     rms_src,
                     self.sample_rate,

rvc/realtime/worker.py

@@ -74,8 +74,10 @@ def _apply_config(vc, cfg):
             vc.crossfade_frame = cf
             vc.extra_frame = ef
             vc.vc_model.realloc(
-                vc.block_frame, vc.extra_frame,
-                vc.crossfade_frame, vc.sola_search_frame,
+                vc.block_frame,
+                vc.extra_frame,
+                vc.crossfade_frame,
+                vc.sola_search_frame,
             )
             vc.generate_strength()
 
@@ -89,6 +91,7 @@ def _apply_config(vc, cfg):
             vc.vc_model.vad = None
         elif vc.vc_model.vad is None:
             from rvc.realtime.utils.vad import VADProcessor
+
             vc.vc_model.vad = VADProcessor(
                 sensitivity_mode=cfg.get("vad_sensitivity", 3),
                 sample_rate=vc.vc_model.sample_rate,
@@ -103,6 +106,7 @@ def _apply_config(vc, cfg):
             strength = cfg.get("clean_strength", 0.5)
             if vc.vc_model.reduced_noise is None:
                 from noisereduce.torchgate import TorchGate
+
                 vc.vc_model.reduced_noise = TorchGate(
                     vc.vc_model.pipeline.tgt_sr,
                     prop_decrease=strength,
@@ -124,6 +128,7 @@ def _apply_config(vc, cfg):
     if model_path and vc.vc_model.model_path != model_path:
         import torch
         import torchaudio.transforms as tat
+
         vc.vc_model.model_path = model_path
         vc.vc_model.pipeline.vc.load_model(model_path)
         vc.vc_model.pipeline.vc.setup_network()
@@ -138,6 +143,7 @@ def _apply_config(vc, cfg):
     sid = cfg.get("sid")
     if sid is not None and vc.vc_model.pipeline.sid != sid:
         import torch
+
         vc.vc_model.pipeline.torch_sid = torch.tensor(
             [sid], device=vc.vc_model.pipeline.device, dtype=torch.int64
         )
@@ -147,9 +153,14 @@ def _apply_config(vc, cfg):
     if index_path is not None:
         if index_path and vc.vc_model.index_path != index_path:
             from rvc.realtime.pipeline import load_faiss_index
+
             index, big_npy = load_faiss_index(
-                index_path.strip().strip('"').strip("\n").strip('"')
-                .strip().replace("trained", "added")
+                index_path.strip()
+                .strip('"')
+                .strip("\n")
+                .strip('"')
+                .strip()
+                .replace("trained", "added")
             )
             vc.vc_model.pipeline.index = index
             vc.vc_model.pipeline.big_npy = big_npy
@@ -174,6 +185,7 @@ def _apply_config(vc, cfg):
         or vc.vc_model.embedder_model_custom != emb_custom
     ):
         from rvc.lib.utils import load_embedding
+
         old = vc.vc_model.pipeline.hubert_model
         del old
         hubert_model = load_embedding(emb, emb_custom)

tabs/realtime/realtime.py

@@ -284,8 +284,13 @@ def refresh_embedders_folders():
 
 
 def save_realtime_settings(
-    input_device, output_device, monitor_device, model_file, index_file,
-    asio_enabled=None, audio_sample_rate=None,
+    input_device,
+    output_device,
+    monitor_device,
+    model_file,
+    index_file,
+    asio_enabled=None,
+    audio_sample_rate=None,
 ):
     """Save realtime settings to config.json"""
     try:
@@ -349,7 +354,9 @@ def load_realtime_settings():
                     "model_file": realtime_config.get("model_file", ""),
                     "index_file": realtime_config.get("index_file", ""),
                     "asio_enabled": realtime_config.get("asio_enabled", False),
-                    "audio_sample_rate": realtime_config.get("audio_sample_rate", 48000),
+                    "audio_sample_rate": realtime_config.get(
+                        "audio_sample_rate", 48000
+                    ),
                 }
     except Exception as e:
         print(f"Error loading realtime settings: {e}")
@@ -532,7 +539,9 @@ def start_realtime(
     _rt_cfg = load_realtime_settings()
     asio_enabled = _rt_cfg["asio_enabled"]
     audio_sample_rate = _rt_cfg["audio_sample_rate"]
-    audio_sample_rate = resolve_sample_rate(input_device_id, asio_enabled, audio_sample_rate)
+    audio_sample_rate = resolve_sample_rate(
+        input_device_id, asio_enabled, audio_sample_rate
+    )
     read_chunk_size = int(chunk_size * audio_sample_rate / 1000 / 128)
 
     callbacks_kwargs = {
@@ -643,7 +652,9 @@ def start_realtime(
                 else 0
             )
             if warmup_remaining > 0:
-                yield i18n("Warming up... ({} blocks remaining)").format(warmup_remaining), interactive_false, interactive_true
+                yield i18n("Warming up... ({} blocks remaining)").format(
+                    warmup_remaining
+                ), interactive_false, interactive_true
             else:
                 yield f"Latency: {audio_manager.latency:.2f} ms | Volume: {audio_manager.volume:.2f} dB", interactive_false, interactive_true
 
@@ -845,11 +856,13 @@ def soundfile_record_audio(
                     now_dir, "assets", "audios", "record_audio.wav"
                 )
 
-            callbacks.vc.send_config({
-                "record_start": True,
-                "record_audio_path": record_audio_path,
-                "export_format": export_format,
-            })
+            callbacks.vc.send_config(
+                {
+                    "record_start": True,
+                    "record_audio_path": record_audio_path,
+                    "export_format": export_format,
+                }
+            )
 
             return "Stop", None
         else:
@@ -1080,7 +1093,6 @@ def realtime_tab():
                         interactive=True,
                     )
 
-
             with gr.TabItem(i18n("Model Settings")):
                 with gr.Row():
                     model_choices = (

tabs/settings/sections/realtime_audio.py

@@ -62,9 +62,7 @@ def realtime_audio_tab():
         with gr.Column():
             asio_enabled = gr.Checkbox(
                 label=i18n("Enable ASIO"),
-                info=i18n(
-                    "Enable ASIO driver support. (Requires restarting Applio)"
-                ),
+                info=i18n("Enable ASIO driver support. (Requires restarting Applio)"),
                 value=saved_asio,
                 interactive=True,
                 visible=IS_WINDOWS,