Update app.py

app.py

@@ -14,15 +14,20 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_french").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)
+def npy_loader(path):
+    np_sample = np.transpose(np.load(path))
+    sample = torch.from_numpy(np_sample)
+    return sample
+#embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+xvector_path = "xvectors/french_recording-bernard-candide_segment_090.npy"
+speaker_embeddings = torch.tensor(npy_loader(xvector_path))
 
 
 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": "transcribe", "language": "fr"})
     return outputs["text"]