Update inference.py

inference.py

@@ -78,11 +78,10 @@ def run_folder(model, args, config, device, verbose: bool = False):
     instruments = prefer_target_instrument(config)[:]
     os.makedirs(args.store_dir, exist_ok=True)
 
-    # Dosya sayısını ve progress için değişkenler
+    # Progress tracking
     total_files = len(mixture_paths)
     current_file = 0
 
-    # Progress tracking
     for path in mixture_paths:
         try:
             # Dosya işleme başlangıcı
@@ -90,76 +89,148 @@ def run_folder(model, args, config, device, verbose: bool = False):
             print(f"Processing file {current_file}/{total_files}")
             
             mix, sr = librosa.load(path, sr=sample_rate, mono=False)
+            mix_orig = mix.copy()
+            
+            if 'normalize' in config.inference:
+                if config.inference['normalize'] is True:
+                    mix, norm_params = normalize_audio(mix)
+
+            # Toplam işlem sürelerini izlemek için başlangıç zamanı
+            total_duration = 0.0
+            total_steps = 100.0  # Toplam %100
+            current_progress = 0.0
+
+            # Model yükleme ve ilk ayrıştırma (%0 -> %30)
+            start_time_step = time.time()
+            waveforms_orig = demix(config, model, mix, device, model_type=args.model_type)
+            step_duration = time.time() - start_time_step
+            total_duration += step_duration
+            current_progress += 30.0 * (step_duration / total_duration) if total_duration > 0 else 30.0
+            print(f"Progress: {min(current_progress, 30.0):.1f}%")
+
+            if args.use_tta:
+                # TTA işlemi (%30 -> %50)
+                start_time_step = time.time()
+                waveforms_orig = apply_tta(config, model, mix, waveforms_orig, device, args.model_type)
+                step_duration = time.time() - start_time_step
+                total_duration += step_duration
+                progress_increment = 20.0 * (step_duration / total_duration) if total_duration > 0 else 20.0
+                for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                    current_progress = min(30.0 + i, 50.0)
+                    time.sleep(0.001)  # Küçük bir gecikme, gerçek işlem için gereksiz olabilir
+                    print(f"Progress: {current_progress:.1f}%")
+
+            if args.demud_phaseremix_inst:
+                print(f"Demudding track (phase remix - instrumental): {path}")
+                instr = 'vocals' if 'vocals' in instruments else instruments[0]
+                instruments.append('instrumental_phaseremix')
+                if 'instrumental' not in instruments and 'Instrumental' not in instruments:
+                    mix_modified = mix_orig - 2*waveforms_orig[instr]
+                    mix_modified_ = mix_modified.copy()
+                    
+                    start_time_step = time.time()
+                    waveforms_modified = demix(config, model, mix_modified, device, model_type=args.model_type)
+                    step_duration = time.time() - start_time_step
+                    total_duration += step_duration
+                    progress_increment = 10.0 * (step_duration / total_duration) if total_duration > 0 else 10.0
+                    for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                        current_progress = min(50.0 + i, 60.0)
+                        time.sleep(0.001)
+                        print(f"Progress: {current_progress:.1f}%")
+
+                    if args.use_tta:
+                        start_time_step = time.time()
+                        waveforms_modified = apply_tta(config, model, mix_modified, waveforms_modified, device, args.model_type)
+                        step_duration = time.time() - start_time_step
+                        total_duration += step_duration
+                        progress_increment = 10.0 * (step_duration / total_duration) if total_duration > 0 else 10.0
+                        for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                            current_progress = min(60.0 + i, 70.0)
+                            time.sleep(0.001)
+                            print(f"Progress: {current_progress:.1f}%")
+                    
+                    waveforms_orig['instrumental_phaseremix'] = mix_orig + waveforms_modified[instr]
+                else:
+                    mix_modified = 2*waveforms_orig[instr] - mix_orig
+                    mix_modified_ = mix_modified.copy()
+                    
+                    start_time_step = time.time()
+                    waveforms_modified = demix(config, model, mix_modified, device, model_type=args.model_type)
+                    step_duration = time.time() - start_time_step
+                    total_duration += step_duration
+                    progress_increment = 10.0 * (step_duration / total_duration) if total_duration > 0 else 10.0
+                    for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                        current_progress = min(50.0 + i, 60.0)
+                        time.sleep(0.001)
+                        print(f"Progress: {current_progress:.1f}%")
+
+                    if args.use_tta:
+                        start_time_step = time.time()
+                        waveforms_modified = apply_tta(config, model, mix_modified, waveforms_orig, device, args.model_type)
+                        step_duration = time.time() - start_time_step
+                        total_duration += step_duration
+                        progress_increment = 10.0 * (step_duration / total_duration) if total_duration > 0 else 10.0
+                        for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                            current_progress = min(60.0 + i, 70.0)
+                            time.sleep(0.001)
+                            print(f"Progress: {current_progress:.1f}%")
+                    
+                    waveforms_orig['instrumental_phaseremix'] = mix_orig + mix_modified_ - waveforms_modified[instr]
+                current_progress = 70.0
+
+            if args.extract_instrumental:
+                instr = 'vocals' if 'vocals' in instruments else instruments[0]
+                waveforms_orig['instrumental'] = mix_orig - waveforms_orig[instr]
+                if 'instrumental' not in instruments:
+                    instruments.append('instrumental')
+
+            # Dosya yazma ve finalize (%70 -> %100)
+            start_time_step = time.time()
+            for instr in instruments:
+                estimates = waveforms_orig[instr]
+                if 'normalize' in config.inference:
+                    if config.inference['normalize'] is True:
+                        estimates = denormalize_audio(estimates, norm_params)
+
+                # Dosya formatı ve PCM türü belirleme
+                is_float = getattr(args, 'export_format', '').startswith('wav FLOAT')
+                codec = 'flac' if getattr(args, 'flac_file', False) else 'wav'
+                
+                # Subtype belirleme
+                if codec == 'flac':
+                    subtype = get_soundfile_subtype(args.pcm_type, is_float)
+                else:
+                    subtype = get_soundfile_subtype('FLOAT', is_float)
+
+                shortened_filename = shorten_filename(os.path.basename(path))
+                output_filename = f"{shortened_filename}_{instr}.{codec}"
+                output_path = os.path.join(args.store_dir, output_filename)
+            
+                sf.write(output_path, estimates.T, sr, subtype=subtype)
+            step_duration = time.time() - start_time_step
+            total_duration += step_duration
+            progress_increment = 20.0 * (step_duration / total_duration) if total_duration > 0 else 20.0
+            for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                current_progress = min(70.0 + i, 90.0)
+                time.sleep(0.001)
+                print(f"Progress: {current_progress:.1f}%")
+
+            # Finalize (%90 -> %100)
+            start_time_step = time.time()
+            time.sleep(0.1)  # Finalize için küçük bir bekleme (gerçek işlem süresiyle değiştirilebilir)
+            step_duration = time.time() - start_time_step
+            total_duration += step_duration
+            progress_increment = 10.0 * (step_duration / total_duration) if total_duration > 0 else 10.0
+            for i in np.arange(0.1, progress_increment + 0.1, 0.1):
+                current_progress = min(90.0 + i, 100.0)
+                time.sleep(0.001)
+                print(f"Progress: {current_progress:.1f}%")
+
         except Exception as e:
             print(f'Cannot read track: {path}')
             print(f'Error message: {str(e)}')
             continue
 
-        mix_orig = mix.copy()
-        if 'normalize' in config.inference:
-            if config.inference['normalize'] is True:
-                mix, norm_params = normalize_audio(mix)
-
-        waveforms_orig = demix(config, model, mix, device, model_type=args.model_type)
-
-        if args.use_tta:
-            waveforms_orig = apply_tta(config, model, mix, waveforms_orig, device, args.model_type)
-
-        if args.demud_phaseremix_inst:
-            print(f"Demudding track (phase remix - instrumental): {path}")
-            instr = 'vocals' if 'vocals' in instruments else instruments[0]
-            instruments.append('instrumental_phaseremix')
-            if 'instrumental' not in instruments and 'Instrumental' not in instruments:
-                mix_modified = mix_orig - 2*waveforms_orig[instr]
-                mix_modified_ = mix_modified.copy()
-                
-                waveforms_modified = demix(config, model, mix_modified, device, model_type=args.model_type)
-                if args.use_tta:
-                    waveforms_modified = apply_tta(config, model, mix_modified, waveforms_modified, device, args.model_type)
-                
-                waveforms_orig['instrumental_phaseremix'] = mix_orig + waveforms_modified[instr]
-            else:
-                mix_modified = 2*waveforms_orig[instr] - mix_orig
-                mix_modified_ = mix_modified.copy()
-                
-                waveforms_modified = demix(config, model, mix_modified, device, model_type=args.model_type)
-                if args.use_tta:
-                    waveforms_modified = apply_tta(config, model, mix_modified, waveforms_orig, device, args.model_type)
-                
-                waveforms_orig['instrumental_phaseremix'] = mix_orig + mix_modified_ - waveforms_modified[instr]
-
-        if args.extract_instrumental:
-            instr = 'vocals' if 'vocals' in instruments else instruments[0]
-            waveforms_orig['instrumental'] = mix_orig - waveforms_orig[instr]
-            if 'instrumental' not in instruments:
-                instruments.append('instrumental')
-
-        for instr in instruments:
-            estimates = waveforms_orig[instr]
-            if 'normalize' in config.inference:
-                if config.inference['normalize'] is True:
-                    estimates = denormalize_audio(estimates, norm_params)
-
-            # Dosya formatı ve PCM türü belirleme
-            is_float = getattr(args, 'export_format', '').startswith('wav FLOAT')
-            codec = 'flac' if getattr(args, 'flac_file', False) else 'wav'
-            
-            # Subtype belirleme
-            if codec == 'flac':
-                subtype = get_soundfile_subtype(args.pcm_type, is_float)
-            else:
-                subtype = get_soundfile_subtype('FLOAT', is_float)
-
-            shortened_filename = shorten_filename(os.path.basename(path))
-            output_filename = f"{shortened_filename}_{instr}.{codec}"
-            output_path = os.path.join(args.store_dir, output_filename)
-        
-            sf.write(output_path, estimates.T, sr, subtype=subtype)
-
-        # Progress yüzdesi hesaplama
-        progress_percent = int((current_file / total_files) * 100)
-        print(f"Progress: {progress_percent}%")
-
     print(f"Elapsed time: {time.time() - start_time:.2f} seconds.")
 
 def proc_folder(args):
@@ -239,4 +310,4 @@ def proc_folder(args):
 
 
 if __name__ == "__main__":
-    proc_folder(None)
+    proc_folder(None)