app.py

import streamlit as st
from models import BagOfModels, SoundToText, TextToSummary
from settings import MODEL_PARSER
args = MODEL_PARSER

st.set_page_config(
    page_title="TTS Applications | Incore Solutions",
    layout="wide",
    menu_items={
        "About": """This is a simple GUI for OpenAI's Whisper.""",
    },
)

def open_instructions():
    with open("instructions.md", "r") as f:
        st.write(f.read())

# Render input type selection on the sidebar & the form
input_type = st.sidebar.selectbox("Input Type", ["YouTube", "File"])

with st.sidebar.form("input_form"):
    if input_type == "YouTube":
        youtube_url = st.text_input("Youtube URL")
    elif input_type == "File":
        input_file = st.file_uploader("File", type=["mp3", "wav"])       

    whisper_model = st.selectbox("Whisper model", options = [whisper for whisper in BagOfModels.get_model_names() if "whisper" in whisper] , index=1) 

    summary = st.checkbox("summarize")
    if summary:
        min_sum = st.number_input("Minimum words in the summary", min_value=1, step=1)
        max_sum = min(min_sum,st.number_input("Maximum words in the summary", min_value=2, step=1))
    st.form_submit_button(label="Save settings")
with st.sidebar.form("save settings"):
    transcribe = st.form_submit_button(label="Transcribe!")
   
   
if transcribe:
    if input_type == "YouTube":
        if youtube_url and youtube_url.startswith("http"):
            model = BagOfModels.load_model(whisper_model,**vars(args))
            st.session_state.transcription = model.predict_stt(source=youtube_url,source_type=input_type,model_task="stt")
        else:
            st.error("Please enter a valid YouTube URL")
            open_instructions()
        
    elif input_type == "File":
        if input_file:
            model = BagOfModels.load_model(whisper_model,**vars(args))
            st.session_state.transcription = model.predict_stt(source=input_file,source_type=input_type,model_task="stt")
        else:
            st.error("Please upload a file")

if "transcription" in st.session_state:
    st.session_state.transcription.whisper()

    # create two columns to separate page and youtube video
    transcription_col, media_col = st.columns(2)
    
    with transcription_col:
        st.markdown("#### Audio")
        with open(st.session_state.transcription.audio_path, "rb") as f:
            st.audio(f.read())
        st.markdown("---")
        st.markdown(f"#### Transcription (whisper model - `{whisper_model}`)")
        st.markdown(f"##### Language: `{st.session_state.transcription.language}`")

        # Trim raw transcribed output off tokens to simplify
        raw_output = st.expander("Raw output")
        raw_output.markdown(st.session_state.transcription.raw_output["text"])

        if summary:
            summarized_output = st.expander("summarized output")
            # CURRENTLY ONLY SUPPORTS 1024 WORD TOKENS -> TODO: FIND METHOD TO INCREASE SUMMARY FOR LONGER VIDS -> 1024 * 4 = aprox 800 words within 1024 range
            text_summary = TextToSummary(str(st.session_state.transcription.text[:1024*4]),min_sum,max_sum).get_summary()
            summarized_output.markdown(text_summary[0]["summary_text"])    

        # Show transcription in format with timers added to text
        time_annotated_output = st.expander("time_annotated_output")
        for segment in st.session_state.transcription.segments:
            time_annotated_output.markdown(
                f"""[{round(segment["start"], 1)} - {round(segment["end"], 1)}] - {segment["text"]}"""
            )
    
    # Show input youtube video
    with media_col:
        if input_type == "YouTube":
            st.markdown("---")
            st.markdown("#### Original YouTube Video")
            st.video(st.session_state.transcription.source)
else:
    pass