import torch
from transformers import Speech2TextProcessor, Speech2TextForConditionalGeneration
from audio_recorder_streamlit import audio_recorder
import numpy as np

def transcribe_audio(audio_bytes):
    
    model = Speech2TextForConditionalGeneration.from_pretrained("facebook/s2t-small-mustc-en-fr-st")
    processor = Speech2TextProcessor.from_pretrained("facebook/s2t-small-mustc-en-fr-st")

    generated_ids = model.generate(input_ids=audio_bytes["input_features"], attention_mask=audio_bytes["attention_mask"])
    translation = processor.batch_decode(generated_ids, skip_special_tokens=True)

return translation


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.")