recognizer.py

from transformers import pipeline#,SpeechT5ForTextToSpeech,SpeechT5Processor,SpeechT5HifiGan
# import torch
# from datasets import load_dataset
# import soundfile as sf
# import matplotlib.pyplot as plt


class Trans:
  def __init__(self,model_name):
    self.model_name=model_name
    self.state=False
    
    

  def initiate(self,**kwargs):
    self.state=True

  def predict(self,input):
    if self.state:
      self.response=self.pipe(input) 
    else:
      raise Exception('Model not initiated') 
      

  def output(self):
    pass


  def process(self):
    pass


class Recognizer(Trans):
  def __init__(self,model_name):
    super().__init__(model_name)
    
    
    

  def initiate(self,**kwargs):
    self.state=True
    self.pipe= pipeline("automatic-speech-recognition",self.model_name,**kwargs)

  # def predict(self,file):
  #   if self.state:
  #     self.response=self.pipe(file) 
  #   else:
  #     raise Exception('Model not initiated') 
      

  def output(self):
    if self.state:
        try:
          return(self.response['text'])
        except AttributeError:
          print('Error: No file was transcribed')
    else:
      raise Exception('Error: Model not initiated') 

  
  def process(self,input):
    self.initiate()
    self.predict(input=input)
    return self.output()

# class Speaker(Trans):
#   def __init__(self,model_name):
#     super().__init__(model_name)
    
    
    

#   def initiate(self,**kwargs):
#     self.state=True
#     self.model=SpeechT5ForTextToSpeech.from_pretrained(self.model_name)
#     self.processor = SpeechT5Processor.from_pretrained(self.model_name)  
#     self.vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

#   def predict(self,text):
#     if self.state:
#       embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
#       self.speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
#       inputs= self.processor(text=text, return_tensors="pt")
#       self.response= self.model.generate_speech(inputs["input_ids"], self.speaker_embeddings, vocoder=self.vocoder)

#     else:
#       raise Exception('Model not initiated') 
      


#   def save(self,file_name="speech.wav"):
#     if self.state:
#         try:
#           sf.write(file_name, self.response.numpy(), samplerate=16000)
#         except AttributeError:
#           print('Error: No text was converted to audio')
#     else:
#       raise Exception('Error: Model not initiated') 
  
  
#   def spectogram(self):
#     plt.figure()
#     plt.imshow(self.response.T)
#     plt.show()


#   def process(self,input,file_name="speech.wav"):
#     self.initiate()
#     self.predict(text=input)
#     self.save(file_name) 

class Translator(Trans):
  def __init__(self,model_name):
    super().__init__(model_name)
    
    
    

  def initiate(self,**kwargs):
    self.state=True
    self.pipe= pipeline("translation",self.model_name,**kwargs)

  

  def output(self):
    if self.state:
        try:
          return(self.response[0]['translation_text'])
        except AttributeError:
          print('Error: No file was transcribed')
    else:
      raise Exception('Error: Model not initiated') 


  def process(self,input):
    self.initiate()
    self.predict(input=input)
    return self.output()  

# r=Recognizer('openai/whisper-tiny.en')
# text_eng=r.process('preamble10.wav')
# print(text_eng)

# t=Translator("Helsinki-NLP/opus-mt-en-fr")
# fre_text=t.process(text_eng)
# print(fre_text)

# s=Speaker("microsoft/speecht5_tts")
# s.process(fre_text)

  
# with open("output.wav", "rb") as f:
#   data = f.read()

# pipe = pipeline("automatic-speech-recognition", "openai/whisper-tiny.en")
# print(pipe('preamble10.wav'))