app.py

import nltk
import librosa
import torch
import gradio as gr
from transformers import WhisperProcessor, WhisperForConditionalGeneration, WhisperTokenizer
nltk.download("punkt")
from transformers import pipeline
import scipy.io.wavfile


model_name = "Shubham09/whisper31filescheck"
processor = WhisperProcessor.from_pretrained(model_name,task="transcribe")
#tokenizer = WhisperTokenizer.from_pretrained(model_name)
model = WhisperForConditionalGeneration.from_pretrained(model_name)

def load_data(input_file):

  #reading the file
  speech, sample_rate = librosa.load(input_file)
  #make it 1-D
  if len(speech.shape) > 1: 
      speech = speech[:,0] + speech[:,1]
  #Resampling the audio at 16KHz
  if sample_rate !=16000:
    speech = librosa.resample(speech, sample_rate,16000)
  return speech
def write_to_file(input_file):
    scipy.io.wavfile.write("microphone-results.wav")
    # with open("microphone-results.wav", "wb") as f:
    #     f.write(input_file.get_wav_data())
    # import base64
    # wav_file = open("temp.wav", "wb")
    # decode_string = base64.b64decode(input_file)
    # wav_file.write(decode_string)  

# def correct_casing(input_sentence):

#   sentences = nltk.sent_tokenize(input_sentence)
#   return (' '.join([s.replace(s[0],s[0].capitalize(),1) for s in sentences]))
pipe = pipeline(model="Shubham09/whisper31filescheck")  # change to "your-username/the-name-you-picked"

def asr_transcript(input_file="microphone-results.wav"):
    text = pipe("microphone-results.wav")["text"]
    return text

  # speech = load_data(input_file)
  # #Tokenize
  # input_features = processor(speech).input_features  #, padding="longest" , return_tensors="pt"
  # #input_values = tokenizer(speech, return_tensors="pt").input_values
  # #Take logits
  # logits = model(input_features).logits
  # #Take argmax
  # predicted_ids = torch.argmax(logits, dim=-1)
  # #Get the words from predicted word ids
  # transcription = processor.batch_decode(predicted_ids)
  # #Correcting the letter casing
  # #transcription = correct_casing(transcription.lower())
  # return transcription

gr.Interface(asr_transcript,
             inputs = gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker"),
             outputs = gr.outputs.Textbox(label="Output Text"),
             title="ASR using Whisper",
             description = "This application displays transcribed text for given audio input",
             examples = [["Actuator.wav"], ["anomalies.wav"]], theme="grass").launch()