app.py

# Transform an audio to text script with language detection.
# Author: Pratiksha Patel
# Description: This script record the audio, transform it to text, detect the language of the file and save it to a txt file.
# import required modules
#import torch
#import streamlit as st
#from audio_recorder_streamlit import audio_recorder
#import numpy as np
#from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq

#def transcribe_audio(audio_bytes):
 #   processor = AutoProcessor.from_pretrained("openai/whisper-large")
  #  model = AutoModelForSpeechSeq2Seq.from_pretrained("openai/whisper-large")
    
    # Convert audio bytes to numpy array
   # audio_array = np.frombuffer(audio_bytes, dtype=np.int16)
    
    # Normalize audio array
    #audio_tensor = torch.tensor(audio_array, dtype=torch.float64) / 32768.0
    
    # Provide inputs to the processor
    ##inputs = processor(audio=audio_tensor, sampling_rate=16000, return_tensors="pt")
    #input_features = processor(audio_tensor, sampling_rate=16000, return_tensors="pt").input_features 

   # generate token ids
    #predicted_ids = model.generate(input_features)
    # decode token ids to text
    #transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)

    #transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
    #return transcription
# Streamlit app
#st.title("Audio to Text Transcription..")

#audio_bytes = audio_recorder(pause_threshold=3.0, sample_rate=16_000)

#if audio_bytes:
 #   st.audio(audio_bytes, format="audio/wav")
    
  #  transcription = transcribe_audio(audio_bytes)

   # if transcription:
    #    st.write("Transcription:")
     #   st.write(transcription)
    #else:
     #   st.write("Error: Failed to transcribe audio.")
#else:
 #   st.write("No audio recorded.")
import torch
import streamlit as st
from audio_recorder_streamlit import audio_recorder
import numpy as np
#from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
# Load model directly
from transformers import AutoProcessor, AutoModelForPreTraining

def transcribe_audio(audio_bytes):
 #   processor = AutoProcessor.from_pretrained("facebook/s2t-small-librispeech-asr")
  #  model = AutoModelForSpeechSeq2Seq.from_pretrained("facebook/s2t-small-librispeech-asr")
    processor = AutoProcessor.from_pretrained("facebook/wav2vec2-base")
    model = AutoModelForPreTraining.from_pretrained("facebook/wav2vec2-base")
    # Convert audio bytes to numpy array
    audio_array = np.frombuffer(audio_bytes, dtype=np.int16)
    # Normalize audio array
    audio_tensor = torch.tensor(audio_array, dtype=torch.float64) / 32768.0
    # Provide inputs to the processor
    input_features = processor(audio_tensor, sampling_rate=16000, return_tensors="pt").input_features
   # generate token ids
    predicted_ids = model.generate(input_features)
    # decode token ids to text
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
    return transcription

# Streamlit app
st.title("Audio to Text Transcription..")

audio_bytes = audio_recorder(pause_threshold=3.0, sample_rate=16_000)
if audio_bytes:
    st.audio(audio_bytes, format="audio/wav")

    transcription = transcribe_audio(audio_bytes)

    if transcription:
        st.write("Transcription:")
        st.write(transcription)
    else:
        st.write("Error: Failed to transcribe audio.")
else:
    st.write("No audio recorded.")