import torch
import whisper
import pytube
import librosa
import streamlit as st
import numpy as np
from fpdf import FPDF
from reportlab.pdfgen.canvas import Canvas
import time



def predict(url=None, translation="No",tran_lang="en"):

    model_m = whisper.load_model("tiny")

    #file_path = 'https://cf-courses-data.s3.ujs.cloud-object-storage.appdomain.cloud/IBM-GPXX0EPMEN/20220627_140242.mp4'
    file_path = 'https://www.youtube.com/watch?v=-WbN61qtTGQ'
    data = pytube.YouTube(file_path)
    speech = data.streams.get_audio_only()
    audio_file = speech.download()
    audio_35 = whisper.load_audio(audio_file)

    audio = whisper.pad_or_trim(audio_35)

    mel = whisper.log_mel_spectrogram(audio).to(model_m.device)

    _,probs = model_m.detect_language(mel)

    p = -1
    for key in probs:
        if probs[key] >p:
            p = probs[key]
    for keys in probs:
        if probs[keys] == p:
            detected_lang = keys

    lang_dict = sorted(probs)
    video_url = url
    v_data = pytube.YouTube(video_url)
    speech = v_data.streams.get_audio_only()
    test_audio_file = speech.download()

    transcription = model_m.transcribe(test_audio_file,fp16=False)["text"]

    if translation == "Yes":
        trans = model_m.transcribe(test_audio_file,language=tran_lang,fp16=False)["text"]
        return detected_lang, transcription, trans
    else:
        return lang_dict, transcription
st.image(image="https://www.respeecher.com/hubfs/What-is-Text-to-Speech-TTS%29-Initial-Speech-Synthesis-Explained-Respeecher-voice-cloning-software.jpeg",output_format="JPEG")
st.title("Sppech to Text generator")

st.write("This app uses an open source neural net called Whisper(developed by OpenAI)")

url = st.text_input(label="Please enter the YouTube url: ")
tran_req = st.selectbox(label="Do you want to translate the transcript?",options=("Yes","No"))

if tran_req=="Yes":
    lang = st.selectbox(label="Please select the required language: ", options=("en","fr","ja"))
else:
    lang = "en"

if st.button("Generate"):
    st.spinner("Fetching the video...")
    lang_d,transcription,trans = predict(url,translation=tran_req,tran_lang=lang)
    
    st.spinner("Speech to Text engine running...")
    time.sleep(1)
    
    st.write("Detected language:",lang_d)
    #canvas = Canvas("transcript.pdf")
    #canvas.drawString(72, 72, transcription)
    #canvas.save()
    #st.download_button(label="Click here to download the transcript", data=canvas, mime='text/csv',file_name="transcript.pdf")
    st.write(transcription)
    st.spinner("Translation in progress..")
    time.sleep(1)
    st.write("Translation: ")
    
    st.write(trans)
    st.spinner("Completed")
    st.success("Speech to text converted successfully!")