import gradio as gr
from transformers import pipeline
from pydub import AudioSegment
import wordtodigits

model = pipeline("automatic-speech-recognition",
                 "facebook/wav2vec2-base-960h")
                 
model2 = gr.Interface.load("huggingface/facebook/fastspeech2-en-ljspeech")


def asr(speech):
  transcript = model(speech)['text']
  strings = transcript.split()
  text = ""
  equation = ""
  symbols = {"plus":"+","minus":"-","times":"*","divide":"/"}
  for i in range(len(strings)):
      if strings[i].lower() in symbols:
          text = wordtodigits.convert(text)
          equation += text + symbols[strings[i].lower()]
          text=""
          continue
      text += strings[i].lower() + " "
      if i == len(strings)-1:
          text = wordtodigits.convert(text)
          equation += text
  
  ans = round(eval(equation),2)
  
  return transcript, equation, ans, model2("The answer is "+str(ans))

gr.Interface(fn=asr,
             #inputs = gr.inputs.Audio(source="microphone", type="filepath", optional=False, label="Please record your voice"),
             inputs = gr.inputs.Audio(source="upload", type="filepath", label="Upload your audio file here"),
			 outputs = [gr.outputs.Textbox(type="str", label="Text Translation"),
						gr.outputs.Textbox(type="str", label="Equation"),
						gr.outputs.Textbox(type="str", label="Answer"),
						gr.outputs.Audio(type="file", label="Speech Answer")],
             title = "Speech Equation Solver",
             description = "This app aims to translate speech into an equation, solve the equation and generate a speech to tell the user the answer to the simple problem",
			 article = "Models: Wav2Vec2-Base-960h, fastspeech2-en-ljspeech",
             examples=["additionTest.mp3","minusTest.mp3","multiplyTest.mp3","divideTest.mp3"]
             ).launch()