Update app.py

app.py

@@ -13,14 +13,26 @@ 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")
+replacements = [
+    ("á", "a"),
+    ("í", "i"),
+    ("ñ", "n"),
+    ("ó", "o"),
+    ("ú", "u"),
+    ("ü", "u"),
+]
 
 print("inside1")
 model = SpeechT5ForTextToSpeech.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_spanish").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_spanish").to(device)
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 print("inside2")
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
+def cleanup_text(text):
+    for src, dst in replacements:
+        text = text.replace(src, dst)
+    return text
 
 def translate(audio):
     outputs = asr_outputs(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "es"})
@@ -28,6 +40,7 @@ def translate(audio):
 
 
 def synthesise(text):
+    text = cleanup_text(text)
     inputs = processor(text=text, return_tensors="pt")
     speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
     return speech.cpu()