init

experiment_voxceleb1.py

@@ -11,17 +11,16 @@ from model_w2v_bert import W2VBertSE
 
 
 cache_dir = p_join("experiment_cache", "voxceleb1")
-voxceleb1_dataset = load_dataset("asahi417/voxceleb1-test-split", split="test")
-print(f"voxceleb1_dataset: {len(voxceleb1_dataset)}")
 
 
-def get_embedding(model_class, model_name: str):
+def get_embedding(model_class, model_name: str, dataset_name: str):
+    dataset = load_dataset(dataset_name, split="test")
     file_path = p_join(cache_dir, f"embedding.{model_name}.json")
     if os.path.exists(file_path):
         return
     model = model_class()
     embeddings = []
-    for i in tqdm(voxceleb1_dataset, total=len(voxceleb1_dataset)):
+    for i in tqdm(dataset, total=len(dataset)):
         start = time()
         v = model.get_speaker_embedding(i["audio"]["array"], i["audio"]["sampling_rate"])
         embeddings.append({
@@ -30,7 +29,8 @@ def get_embedding(model_class, model_name: str):
             "sampling_rate": i["audio"]["sampling_rate"],
             "id": i["id"],
             "speaker_id": i["speaker_id"],
-            "process_time": time() - start
+            "process_time": time() - start,
+            "dataset_name": os.path.basename(dataset_name)
         })
     with open(file_path, "w") as f:
         f.write("\n".join([json.dumps(i) for i in embeddings]))
@@ -38,8 +38,8 @@ def get_embedding(model_class, model_name: str):
 
 if __name__ == '__main__':
     # cache embedding
-    get_embedding(MetaVoiceSE, "meta_voice_se")
-    get_embedding(PyannoteSE, "pyannote_se")
-    get_embedding(W2VBertSE, "w2v_bert_se")
+    get_embedding(MetaVoiceSE, "meta_voice_se", "asahi417/voxceleb1-test-split")
+    get_embedding(PyannoteSE, "pyannote_se", "asahi417/voxceleb1-test-split")
+    get_embedding(W2VBertSE, "w2v_bert_se", "asahi417/voxceleb1-test-split")
 
 

model_meta_voice.py

@@ -14,8 +14,8 @@ import torch
 from torch import nn
 
 
-checkpoint_url = "https://huggingface.co/datasets/asahi417/experiment-speaker-embedding/resolve/main/meta_voice_speaker_encoder.pt.pt"
-model_weight = p_join(os.path.expanduser('~'), ".cache", "experiment_speaker_embedding", "meta_voice_speaker_encoder.pt.pt")
+checkpoint_url = "https://huggingface.co/datasets/asahi417/experiment-speaker-embedding/resolve/main/meta_voice_speaker_encoder.pt"
+model_weight = p_join(os.path.expanduser('~'), ".cache", "experiment_speaker_embedding", "meta_voice_speaker_encoder.pt")
 
 
 def wget(url: str, output_file: Optional[str] = None):