Change output language to spanish.

app.py

@@ -3,40 +3,38 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-
+from transformers import pipeline
+from transformers import VitsModel, AutoTokenizer
+#import torch
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
 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")
-
-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)
+# load text-to-speech checkpoint
+model = VitsModel.from_pretrained("facebook/mms-tts-spa")
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-spa")
 
 
 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":"es"})
     return outputs["text"]
 
 
 def synthesise(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()
+    inputs = tokenizer(text, return_tensors="pt")
+
+    with torch.no_grad():
+        speech = model(**inputs).waveform
+    return speech[0].cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
-    return 16000, synthesised_speech
+    return model.config.sampling_rate, synthesised_speech
 
 
 title = "Cascaded STST"
@@ -51,7 +49,7 @@ demo = gr.Blocks()
 
 mic_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="microphone", type="filepath"),
+    inputs=gr.Audio(sources=["microphone"], type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     title=title,
     description=description,
@@ -59,7 +57,7 @@ mic_translate = gr.Interface(
 
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="upload", type="filepath"),
+    inputs=gr.Audio(sources=["upload"], type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     examples=[["./example.wav"]],
     title=title,