Update app.py

app.py

@@ -12,17 +12,41 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+# processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+# model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
+vocoder = SpeechT5HifiGan.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
 
+# embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+# speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+
+dataset_nl = load_dataset("facebook/voxpopuli", "nl", split="train", streaming=True)
+data_list = []
+speaker_embeddings_list = []
+
+for i, data in enumerate(iter(dataset_nl)):
+    # print(i)
+    if(i > 5):
+      break
+    data_list.append(data)
+    # data = next(iter(dataset_nl))
+    text = data["raw_text"]
+    # print(data)
+    speaker_embeddings = create_speaker_embedding(data["audio"]["array"])
+    speaker_embeddings = torch.tensor(speaker_embeddings)[None]
+    speaker_embeddings_list.append(speaker_embeddings)
+
+
+speaker_embeddings = speaker_embeddings_list[4]
+    
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    # outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    outputs = pipe(audio, max_new_tokens=256, generate_kwargs={"language":"<|nl|>","task": "transcribe"})
     return outputs["text"]