Update app.py

app.py

@@ -2,64 +2,73 @@ import gradio as gr
 import numpy as np
 import torch
 from datasets import load_dataset
-
 from transformers import pipeline
-from transformers import VitsModel, VitsTokenizer
-from transformers import SpeechT5Processor, SpeechT5Processor, SpeechT5HifiGan
-
 
 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)
 
+
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["text"]
 
+'''
+from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
+
+# 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)
+
+
+def synthesise_old(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()
+
+
+def speech_to_speech_translation_old(audio):
+    translated_text = translate(audio)
+    synthesised_speech = synthesise_old(translated_text)
+    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
+    return 16000, synthesised_speech
+'''
+
+from transformers import VitsModel, VitsTokenizer
+
 
 # load translator to french
 en_fr_translator = pipeline("translation_en_to_fr")
 
-
 # load text-to-speech
 model_new = VitsModel.from_pretrained("facebook/mms-tts-fra")
 tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-fra")
 
-def synthesise_new(text):
+
+def synthesise(text):
     translation_to_french = en_fr_translator(text)
     french_text = translation_to_french[0]['translation_text']
 
     inputs = tokenizer(french_text, return_tensors="pt")
     input_ids = inputs["input_ids"]
-    
+
     with torch.no_grad():
         outputs = model_new(input_ids)
-    
+
     speech = outputs["waveform"]
     return speech
 
 
-# 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)
-
-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()
-
-
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
-    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
+    synthesised_speech = (synthesised_speech[0].numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech