app.py

import streamlit as st
import torch
from transformers import pipeline
import soundfile as sf
import io
import numpy as np
from datetime import timedelta

# Set page configuration
st.set_page_config(
    page_title="Audio Transcription with Whisper",
    page_icon="🎙️",
    layout="wide"
)

# Initialize speech recognition pipeline
@st.cache_resource
def load_model():
    device = "cuda:0" if torch.cuda.is_available() else "cpu"
    pipe = pipeline(
        "automatic-speech-recognition",
        model="openai/whisper-small",
        chunk_length_s=30,
        device=device,
    )
    return pipe

def format_timestamp(seconds):
    """Convert seconds to HH:MM:SS format"""
    return str(timedelta(seconds=seconds)).split('.')[0]

def process_audio(audio_file, return_timestamps=False):
    """Process audio file and return transcription"""
    try:
        # Read audio file
        audio_bytes = audio_file.read()
        audio_array, sampling_rate = sf.read(io.BytesIO(audio_bytes))
        
        # Convert to mono if stereo
        if len(audio_array.shape) > 1:
            audio_array = audio_array.mean(axis=1)
        
        # Get predictions
        pipe = load_model()
        if return_timestamps:
            result = pipe({"sampling_rate": sampling_rate, "raw": audio_array}, 
                        batch_size=8, return_timestamps=True)
            return result["chunks"]
        else:
            result = pipe({"sampling_rate": sampling_rate, "raw": audio_array}, 
                        batch_size=8)
            return result["text"]
            
    except Exception as e:
        return f"Error processing audio: {str(e)}"

# Sidebar
with st.sidebar:
    st.header("Settings")
    show_timestamps = st.checkbox("Show timestamps", value=False)
    
    st.markdown("---")
    st.markdown("### Model Information")
    st.markdown("""
    - Model: OpenAI Whisper (small)
    - Device: {device}
    - Max chunk length: 30 seconds
    """.format(device="GPU" if torch.cuda.is_available() else "CPU"))
    
    st.markdown("---")
    st.markdown("### Supported Audio Formats")
    st.markdown("""
    - WAV
    - FLAC
    - MP3
    - OGG
    """)

# Main app
st.title("🎙️ Audio Transcription with Whisper")
st.markdown("Upload an audio file to get its transcription")

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

if audio_file is not None:
    # Create two columns
    col1, col2 = st.columns([1, 1])
    
    with col1:
        st.subheader("Audio File")
        st.audio(audio_file)
        
        file_details = {
            "Filename": audio_file.name,
            "File size": f"{audio_file.size / 1024:.2f} KB",
            "File type": audio_file.type
        }
        
        st.json(file_details)
    
    with col2:
        st.subheader("Transcription")
        
        if st.button("Start Transcription"):
            with st.spinner("Processing audio..."):
                if show_timestamps:
                    results = process_audio(audio_file, return_timestamps=True)
                    
                    if isinstance(results, list):
                        for chunk in results:
                            st.markdown(f"""
                            **[{format_timestamp(chunk['timestamp'][0])} - {format_timestamp(chunk['timestamp'][1])}]**  
                            {chunk['text']}
                            """)
                    else:
                        st.error(results)
                else:
                    transcription = process_audio(audio_file, return_timestamps=False)
                    if not transcription.startswith("Error"):
                        st.write(transcription)
                    else:
                        st.error(transcription)

# Download transcription
if 'transcription' in locals():
    if not transcription.startswith("Error"):
        st.download_button(
            label="Download Transcription",
            data=transcription,
            file_name=f"{audio_file.name}_transcription.txt",
            mime="text/plain"
        )

# Footer
st.markdown("---")
st.markdown("""
Built with:
- Streamlit
- OpenAI Whisper
- 🤗 Transformers
""")

# Display warnings/info
st.sidebar.markdown("---")
if not torch.cuda.is_available():
    st.sidebar.warning("⚠️ Running on CPU. Processing might be slower.")