Fix dataset handling with the new embedding file keys (#1991)

TTS/bin/compute_embeddings.py

@@ -102,11 +102,9 @@ speaker_mapping = {}
 for idx, fields in enumerate(tqdm(samples)):
     class_name = fields[class_name_key]
     audio_file = fields["audio_file"]
-    dataset_name = fields["dataset_name"]
+    embedding_key = fields["audio_unique_name"]
     root_path = fields["root_path"]
 
-    relfilepath = os.path.splitext(audio_file.replace(root_path, ""))[0]
-    embedding_key = f"{dataset_name}#{relfilepath}"
     if args.old_file is not None and embedding_key in encoder_manager.clip_ids:
         # get the embedding from the old file
         embedd = encoder_manager.get_embedding_by_clip(embedding_key)

TTS/tts/datasets/__init__.py

@@ -59,6 +59,18 @@ def split_dataset(items, eval_split_max_size=None, eval_split_size=0.01):
     return items[:eval_split_size], items[eval_split_size:]
 
 
+def add_extra_keys(metadata, language, dataset_name):
+    for item in metadata:
+        # add language name
+        item["language"] = language
+        # add unique audio name
+        relfilepath = os.path.splitext(item["audio_file"].replace(item["root_path"], ""))[0]
+        audio_unique_name = f"{dataset_name}#{relfilepath}"
+        item["audio_unique_name"] = audio_unique_name
+
+    return metadata
+
+
 def load_tts_samples(
     datasets: Union[List[Dict], Dict],
     eval_split=True,
@@ -111,15 +123,15 @@ def load_tts_samples(
         # load train set
         meta_data_train = formatter(root_path, meta_file_train, ignored_speakers=ignored_speakers)
         assert len(meta_data_train) > 0, f" [!] No training samples found in {root_path}/{meta_file_train}"
-        meta_data_train = [{**item, **{"language": language, "dataset_name": dataset_name}} for item in meta_data_train]
+
+        meta_data_train = add_extra_keys(meta_data_train, language, dataset_name)
+
         print(f" | > Found {len(meta_data_train)} files in {Path(root_path).resolve()}")
         # load evaluation split if set
         if eval_split:
             if meta_file_val:
                 meta_data_eval = formatter(root_path, meta_file_val, ignored_speakers=ignored_speakers)
-                meta_data_eval = [
+                meta_data_eval = add_extra_keys(meta_data_eval, language, dataset_name)
-                    {**item, **{"language": language, "dataset_name": dataset_name}} for item in meta_data_eval
-                ]
             else:
                 meta_data_eval, meta_data_train = split_dataset(meta_data_train, eval_split_max_size, eval_split_size)
             meta_data_eval_all += meta_data_eval

TTS/tts/datasets/dataset.py

@@ -256,6 +256,7 @@ class TTSDataset(Dataset):
             "speaker_name": item["speaker_name"],
             "language_name": item["language"],
             "wav_file_name": os.path.basename(item["audio_file"]),
+            "audio_unique_name": item["audio_unique_name"],
         }
         return sample
 
@@ -397,8 +398,8 @@ class TTSDataset(Dataset):
                 language_ids = None
             # get pre-computed d-vectors
             if self.d_vector_mapping is not None:
-                wav_files_names = list(batch["wav_file_name"])
+                embedding_keys = list(batch["audio_unique_name"])
-                d_vectors = [self.d_vector_mapping[w]["embedding"] for w in wav_files_names]
+                d_vectors = [self.d_vector_mapping[w]["embedding"] for w in embedding_keys]
             else:
                 d_vectors = None
 

TTS/tts/models/vits.py

@@ -284,6 +284,7 @@ class VitsDataset(TTSDataset):
             "wav_file": wav_filename,
             "speaker_name": item["speaker_name"],
             "language_name": item["language"],
+            "audio_unique_name": item["audio_unique_name"],
         }
 
     @property
@@ -308,6 +309,7 @@ class VitsDataset(TTSDataset):
             - language_names: :math:`[B]`
             - audiofile_paths: :math:`[B]`
             - raw_texts: :math:`[B]`
+            - audio_unique_names: :math:`[B]`
         """
         # convert list of dicts to dict of lists
         B = len(batch)
@@ -348,6 +350,7 @@ class VitsDataset(TTSDataset):
             "language_names": batch["language_name"],
             "audio_files": batch["wav_file"],
             "raw_text": batch["raw_text"],
+            "audio_unique_names": batch["audio_unique_name"],
         }
 
 
@@ -1470,7 +1473,7 @@ class Vits(BaseTTS):
         # get d_vectors from audio file names
         if self.speaker_manager is not None and self.speaker_manager.embeddings and self.args.use_d_vector_file:
             d_vector_mapping = self.speaker_manager.embeddings
-            d_vectors = [d_vector_mapping[w]["embedding"] for w in batch["audio_files"]]
+            d_vectors = [d_vector_mapping[w]["embedding"] for w in batch["audio_unique_names"]]
             d_vectors = torch.FloatTensor(d_vectors)
 
         # get language ids from language names

tests/data/ljspeech/speakers.json

@@ -1,5 +1,5 @@
 {
-    "LJ001-0001.wav": {
+    "#/wavs/LJ001-0001": {
         "name": "ljspeech-0",
         "embedding": [
             0.05539746582508087,
@@ -260,7 +260,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0002.wav": {
+    "#/wavs/LJ001-0002": {
         "name": "ljspeech-1",
         "embedding": [
             0.05539746582508087,
@@ -521,7 +521,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0003.wav": {
+    "#/wavs/LJ001-0003": {
         "name": "ljspeech-2",
         "embedding": [
             0.05539746582508087,
@@ -782,7 +782,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0004.wav": {
+    "#/wavs/LJ001-0004": {
         "name": "ljspeech-3",
         "embedding": [
             0.05539746582508087,
@@ -1043,7 +1043,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0005.wav": {
+    "#/wavs/LJ001-0005": {
         "name": "ljspeech-4",
         "embedding": [
             0.05539746582508087,
@@ -1304,7 +1304,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0006.wav": {
+    "#/wavs/LJ001-0006": {
         "name": "ljspeech-5",
         "embedding": [
             0.05539746582508087,
@@ -1565,7 +1565,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0007.wav": {
+    "#/wavs/LJ001-0007": {
         "name": "ljspeech-6",
         "embedding": [
             0.05539746582508087,
@@ -1826,7 +1826,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0008.wav": {
+    "#/wavs/LJ001-0008": {
         "name": "ljspeech-7",
         "embedding": [
             0.05539746582508087,
@@ -2087,7 +2087,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0009.wav": {
+    "#/wavs/LJ001-0009": {
         "name": "ljspeech-8",
         "embedding": [
             0.05539746582508087,
@@ -2348,7 +2348,7 @@
             -0.09469571709632874
         ]
     },
-    "LJ001-0010.wav": {
+    "#/wavs/LJ001-0010": {
         "name": "ljspeech-9",
         "embedding": [
             0.05539746582508087,