UPDAED app.py with català speaker embedding

app.py

@@ -1,9 +1,11 @@
+import os
+import torch
 import gradio as gr
 import numpy as np
 import torch
 from datasets import load_dataset
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-
+from speechbrain.pretrained import EncoderClassifier
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
@@ -18,12 +20,59 @@ model = SpeechT5ForTextToSpeech.from_pretrained(
     "JanLilan/speecht5_finetuned_openslr-slr69-cat"
 ).to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+
+######################################################################################
+################################## SPEAKER EMBEDDING #################################
+######################################################################################
 # we will try to translate with this voice embedding... Let's see what happen. else:
-# dataset = load_dataset("projecte-aina/openslr-slr69-ca-trimmed-denoised", split="train")
-# dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
+dataset = load_dataset("projecte-aina/openslr-slr69-ca-trimmed-denoised", split="train")
+dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
+# LOAD
+spk_model_name = "speechbrain/spkrec-xvect-voxceleb"
+speaker_model = EncoderClassifier.from_hparams(
+    source=spk_model_name,
+    run_opts={"device": device},
+    savedir=os.path.join("/tmp", spk_model_name),
+)
+
+def create_speaker_embedding(waveform):
+    with torch.no_grad():
+        speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform))
+        speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=2)
+        speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
+    return speaker_embeddings
+
+# we must take one speaker embeding
+checkpoint = "microsoft/speecht5_tts"
+processor = SpeechT5Processor.from_pretrained(checkpoint)
+
+# function to embedd
+def prepare_dataset(example):
+    audio = example["audio"]
+
+    example = processor(
+        text=example["transcription"],
+        audio_target=audio["array"],
+        sampling_rate=audio["sampling_rate"],
+        return_attention_mask=False,
+    )
+
+    # strip off the batch dimension
+    example["labels"] = example["labels"][0]
+
+    # use SpeechBrain to obtain x-vector
+    example["speaker_embeddings"] = create_speaker_embedding(audio["array"])
+
+    return example
+
+processed_example = prepare_dataset(dataset[0])
+speaker_embeddings = torch.tensor(processed_example["speaker_embeddings"]).unsqueeze(0)
+
+
+
 # etc.
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+# embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+# speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
 
 def translate(audio):