app.py

import streamlit as st
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import torchaudio

# Load the model
@st.cache_resource
def load_model():
    processor = WhisperProcessor.from_pretrained("233-Yorozuya/dl_whisper_model")
    model = WhisperForConditionalGeneration.from_pretrained("233-Yorozuya/dl_whisper_model")
    return processor, model

processor, model = load_model()

st.title("ASR with Fine-Tuned Whisper")
st.write("Upload an audio file for transcription:")

# File upload
audio_file = st.file_uploader("Choose an audio file", type=["wav", "mp3", "ogg"])

if audio_file:
    # Load and preprocess audio
    audio, rate = torchaudio.load(audio_file)
    audio = torchaudio.transforms.Resample(orig_freq=rate, new_freq=16000)(audio)
    inputs = processor(audio[0].numpy(), sampling_rate=16000, return_tensors="pt")

    # Perform inference
    with st.spinner("Transcribing..."):
        predicted_ids = model.generate(inputs.input_features)
        transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]

    # Display result
    st.subheader("Transcription")
    st.write(transcription)