Fix bugs for content extractor's args (#165)

Author: RMSnow

config/svc/base.json

@@ -86,4 +86,34 @@
         "spk2id": "singers.json",
         "utt2spk": "utt2singer"
     },
+    "model": {
+        "condition_encoder": {
+            "merge_mode": "add",
+            // Prosody Features
+            "use_f0": true,
+            "use_uv": true,
+            "use_energy": true,
+            // Quantization (0 for not quantization)
+            "input_melody_dim": 1,
+            "n_bins_melody": 256,
+            "output_melody_dim": 384,
+            "input_loudness_dim": 1,
+            "n_bins_loudness": 256,
+            "output_loudness_dim": 384,
+            // Semantic Features
+            "use_whisper": false,
+            "use_contentvec": false,
+            "use_wenet": false,
+            "use_mert": false,
+            "whisper_dim": 1024,
+            "contentvec_dim": 256,
+            "mert_dim": 256,
+            "wenet_dim": 512,
+            "content_encoder_dim": 384,
+            // Speaker Features
+            "output_singer_dim": 384,
+            "singer_table_size": 512,
+            "use_spkid": true
+        }
+    },
 }

processors/content_extractor.py

@@ -526,10 +526,7 @@ def extract_utt_content_features_dataloader(cfg, metadata, num_workers):
                 for batch_idx, items in enumerate(tqdm(data_loader)):
                     _metadata, wavs, lens = items
 
-                    batch_content_features = extractor.extract_content_features(
-                        wavs,
-                        lens,
-                    )
+                    batch_content_features = extractor.extract_content_features(wavs)
                     for index, utt in enumerate(_metadata):
                         extractor.save_feature(utt, batch_content_features[index])
 
@@ -561,9 +558,7 @@ def extract_utt_content_features_dataloader(cfg, metadata, num_workers):
                 for batch_idx, items in enumerate(tqdm(data_loader)):
                     _metadata, wavs, lens = items
 
-                    batch_content_features = extractor.extract_content_features(
-                        wavs, lens
-                    )
+                    batch_content_features = extractor.extract_content_features(wavs)
                     for index, utt in enumerate(_metadata):
                         extractor.save_feature(utt, batch_content_features[index])
 
@@ -626,9 +621,6 @@ def extract_utt_content_features_dataloader(cfg, metadata, num_workers):
                 for batch_idx, items in enumerate(tqdm(data_loader)):
                     _metadata, wavs, lens = items
 
-                    batch_content_features = extractor.extract_content_features(
-                        wavs,
-                        lens,
-                    )
+                    batch_content_features = extractor.extract_content_features(wavs)
                     for index, utt in enumerate(_metadata):
                         extractor.save_feature(utt, batch_content_features[index])