demo

app.py

@@ -24,6 +24,23 @@ speaker_model = EncoderClassifier.from_hparams(
     savedir=os.path.join("/tmp", spk_model_name),
 )
 
+def prepare_dataset(examp):
+    audio = examp["audio"]
+
+    examp = processor(
+        text=examp["sentence"],
+        audio_target=audio["array"],
+        sampling_rate=audio["sampling_rate"],
+        return_attention_mask=False,
+    )
+
+    # strip off the batch dimension
+    examp["labels"] = examp["labels"][0]
+
+    # use SpeechBrain to obtain x-vector
+    examp["speaker_embeddings"] = create_speaker_embedding(audio["array"])
+
+    return examp
 
 def create_speaker_embedding(waveform):
     with torch.no_grad():