Generating Transcripts and News Article with Whisper, GPT-3.5, ChatGPT and Streamlit

Key Takeaways

hello everyone welcome to AI anytime channel so today we will work on a very interesting project we will develop an AI powered application using open ai's gpt3 models and Whisper model to you know do speech to text and then we'll use extremely to build the UI and integrate the model into it okay so what we are going to do into this video is that you know we'll take a very uh small YouTube video short short YouTube video and then we'll generate the news article you know based on that video so first we will get the audio from that video using pi tube because it's a YouTube video we can use pythube library and I'll be covering that in bit and once we have the audio file we'll send it to whisper we use the whisper model locally and then we'll get the transcript once we have the transcript we'll use that transcript to get the news article we'll do prompt engineering uh with the help of DaVinci 3 Model by open Ai and then we'll get the news article and the news article and transcript will have options to download them so end user can download both the files as well from extremely directly so this this can be very powerful when you know the end user that can be journalist or some you know any media houses they can use this kind of application they still use it but with open AI so being so Advanced you know this can also improve the process right now so they can add their own thoughts they can do some research and validation and they can create a very powerful news article based on the audio or video recordings okay so let's see how we can do that so we'll we'll be again using chat GPT as a co-founder we'll ask Chi GPT that what the right approach to you know uh perform this task let's see that so I'll be doing a prompt engineering here again I'll say I am developing uh an AI powered application in extremely where I will generate excuse me I'll generate the news article from the video uh for video I'll use Pi tube to get the YouTube video and then whisper to do is speech to text to get the transcript to get the transcript in end I will feed this transcript to DaVinci 3 by open AI to get the news article can you help me with the help me with the 5 Steps approach to perform this task so this is my query that I have asked to share GPT now charge if it is you know going to give me a an approach that how I can build this solution right and it's going to be in five steps the first step is to you know use Pi tube which is a python Library you know to work with YouTube videos okay we'll use this library to you know perform our operations on the video that we are getting from YouTube so use Pi tube to download the YouTube video then convert the video into audio will not do that will directly use Pi tube functions to only download the audio file okay we'll say only audio equal to true I'll be covering that what I'm talking about and then once we have the audio will you we use whisper model to perform speech to text we'll extract the transcript out of that audio file once we have the transcript we will do the prompt engineering okay with the help of open AI API and we'll be using DaVinci 3 which is the most capable model of open AI at this moment which is also be called as GPT 3.5 and in end we know we'll then we'll have the option to download both the transcript and the uh news article so you can also download this you know uh this repository the code base is here on the AI anytime official GitHub repository you know you can download it from here and this is a video we are going to use let me just play this video it's a Trump's interview some some news reporter or some journalists taking interview you know of trump and he's talking about you know on Muslims okay and we'll use this and then on base of this video we will generate the news article let's let's see what he is talking about okay that's what he said okay are they a problem you tell me and I say I don't see the people and I use it often that knock down the World Trade Center going back to Sweden so he said Muslims are about that was part of what he said can you make the case that it's been taught and by the way most Muslims I know many are great people just so you put it on the record I think a certain segment are certainly a problem you want to be so politically correct when you want me to say oh absolutely not I mean you have ice so this is the video that we are going to take it's a very short video around you know two minutes we'll use this video we'll get the a transcript from this video and then we'll generate the news article you know on top of this video so let's see how we can do that okay so what is going to be our step okay so let me just go to tab and I'll be covering that okay so so this is going to be the video we'll use Pi tube this is a python Library to get the audio so we are getting the audio here video from YT which is the YouTube and once we have the audio now we'll use whisper which is the state of the art you know ASR model as released last year by open AI which is an open source model you can fine tune on your own custom data if you have okay whisper and will get the uh will perform this stt which is again the transcript so we'll get the transcript I don't know why I did these vectors it doesn't make any sense but anyway we we have uh audio and then we perform the speech to text to get the transcript once we have the transcript we'll pass this to you know we'll do our prompt engineering and we'll pass it to DaVinci 3 open AI model and you know then we will get the news article so this all will be wrapped you know inside the UI will use extremely which will have a file a YouTube link option as a input will give the YouTube link as an input and then we'll perform all these operations you see over here right we have video and then we'll use this libraries to get the audio sir input will be video and the output will be a news article that will be a txt file that we'll do so let's see how we can build this let me bring my monitor up Okay so let's go to vs code let me open vs code open in terminal okay so these are the requirements that we have the dependencies or the libraries to you know build this application we need streamlit because we are going to use the stream late you know to integrate the models and you know represent you know very good UI very minimal UI and then we'll use Pi tube uh to you know get the YouTube video and we'll be using open AI whisper you know to get the transcript and then we'll use open AI to perform the prompt engineer engineering and get the news article using uh open AI model like DaVinci 3 and then we have python dot Envy where we have to use an EnV file or Json file or some file to you know store our credentials of open AI the API key so what I will do now I'll create an app.pi here and here I will write all my code so guys first thing that we have to you know import is the streamlit because we are going to use extremely here right so let's let's start importing the libraries and modules so streamlit as HD and if you haven't installed this Library yet you have to create this requirements.txt and you can also get it from the GitHub uh go and take the reposit Clone the repository and then install these dependencies at extremely Pi tube open AIS for open air in Python dot EnV and then once we installed all those things we first that we have imported the import stimuli type HD now what we have to do we have to import open AI we once we import open a we'll do import Pi tube Pi tip to you know work with the YouTube videos so what I will do I will do PI tube import Q2 yeah so now what we do we also import OS to get the keys and all and then from quietly import path excuse me input path okay and then we can also import executive and then we need to import dot ER load dot EnV import load dot EnV Y and then we have to import whisper and then we have to import what else gif file because we will have this option where end user can you know download this transcript and the news article you know in a zip file okay all together first thing I will do I will say load dot EnV which is the function and now what I will Define my open AI API key variable here which will import the my credentials from that EnV file that I have created you can see I have a EMB file here right dot EnV where you can store your you know sensitive credentials like the API key is one of them okay OS dot get EnV and he'll pass uh the variable that you have you know created in the EnV file where you are storing it okay you have to create a variable and then it's a string right and that's the variable you have to mention here which is open AI underscore API key in my case now what we will do we will use whisper model okay so whisper has around five models if I am not wrong they have five open source model uh tiny small a tiny base small medium and large okay that depends on you know different weights the parameters that model has been trained on okay and tiniest being tiny being the smallest one and large being the are they gonna biggest one okay that depends if you have GPU machine if you have it's totally depends on what kind of computational power you have in your machine to run those models so today I will go with you know base model okay which comes after tiny because I think this is just uh we're just creating a video here we're not building a product you know depending on the computational power if you have if you have any Nvidia GPU or even on cloud if you are using some kind of GPU you can use any other models and you can try it so what I will do is write the function you know to load this model so I'm creating a function called load model here here I will write code to load the model I'll say model equals V Square dot input we have imported whisper above we have installed pip install open AI hyphen whisper and now what I will do I will say load model and then I just pass which is a string which will be base here so this is the model that I'm import loading okay base and I'll just return the model and I will use this model to get the transcript out of the audio file will also will also perform a decorator here okay that's called hd.cash to keep the model so model will not be every time I'm running the code for changing something in the code or when I'm doing some testing it should not load the model every time that's what HCL provide that we will mention later now let's start writing the YouTube video part okay where we will get the audio from the YouTube video URL directly let's see how we can do that so what I will do I'll say Define and I also I'll also save this audio you know in directory so I'm calling it save audio and here I'll give the URL so I'm passing URL as an parameter here that will be a parameter to this function which will take YouTube URL and it will give you the audio it will automatically saved as well now let's see Define a variable called YT and this is My URL so we have if you can see over here we have imported U2 and that's what we are using this Method YouTube and passing this parameter URL now first we have to download this video and then we have to you know only get the audio so YT dot streams you see it's suggesting me here why dot streams Dot uh filter so Pi tube you know you can when you have imported this uh Library by 2 you can do uh you can check what all the modules available inside it how do we do it we do dir which is directory and then pass the input that you have now YouTube and it will give you all the modules inside it okay so filter now what I will do here guys I will you can see there are two parameters in C there are two parents right which is very famous which has been shown here only Audio Only video what I want to do I only want to get the Audio I don't want to get the you know in this case I can only get the audio from here okay and then first very correct okay I'm only getting the audio from this URL which is an YouTube video URL okay and now my output files I'll say out file equals let me download so I will download this variable video dot download it's a function again that's it so we have an output file now what we'll do we only get we do not want to get the entire path name of the video that we are getting it or the audio we are getting it from YouTube what we'll do we'll only split the let's see how we can do split text and out file and now let's uh handle the file name so what we have to do uh let me just show you what I'm talking about the file name base plus we're defining it the extension here that we only need the MP3 file MP3 you do not need anything else okay we need only the MP3 file now we have the file name now what we can do but what if there is nothing available right if if you are if you are facing some challenges to you know uh suppose you have given a YouTube url YouTube video URL and it's not working for your program how will you handle this basically the error how will you do the error handling here what we can do we can use try cash to see so try OS dot rename you're trying to rename so as dot rename uh out file and the file name yeah and then we will write the accept what if there's an error so what will it call Windows error Windows error okay it's not a function except window error except Windows error and Os Dot remove will remove the file name it's very good it's suggesting me automatically OS dot remove and then OS Dot rename so basically what we are doing here is if you are checking if the file is already inside you know downloaded the same file name we will not download it okay if there's if if there's on we can also replace it right so always dot remove file name and always dot rename so we can also rename it okay except we get any not getting any Windows error okay then this program will be executed okay except Windows error okay now let's Define a variable called audio file name is so what we are going to do here we will if we have download you will see it when we download and you know video basically from YouTube url link and when we are saving the audio the file name is kind of very lengthy okay and we do not need that complete file name so what I will do I'll just say file name.stem and then plus dot MP3 yeah this looks good plus HTM dot MP3 audio file name and now let's print that in terminal and C so print YT dot or title let's print the title we do not need this multi.title and let's print a message has successfully has successfully been downloaded as a downloaded okay that's such like this okay as downloaded yeah print YT title plus SSS will be known this will print it in the terminal and we'll see if we are you know if you are able to get the desired response that we need let's now print the file name so five name and now let's return this uh YT dot title and audio dot character and audio file name so this is the uh YouTube video function that we have you know we have written over here okay what we will do we'll run this program now but how we run this program now let me do one thing okay so save audio let's also write the all the other other functions that we have and then we'll test it out so we have save video audio URL and we have downloaded this audio okay now what we can do we can also write the function for audio to transcript now we have the audio file name here okay and it will not take much time because it's got couple of lines of code and we'll be able to you know get the transcript using whisper so what we'll do now we'll do save audio to transcript and here we'll pass the audio file so the audio file will be an input and then we'll use the model here right we have to use this base model that we have created the function to load the model okay so what we will do we'll say model and we'll use this function now we'll say load model so here we are loading the model and now we'll just write a result and from here we are getting the audio so what we'll do we'll say model dot transcribe model.transcribe and here we'll pass this model transcribe audio file now we have the result okay so what we'll do we'll say transcript from script our transcript will be a result and the text if you only need the text we do not need the other metadata or something that we are getting when we are running this base model or any other model from whisper and we'll just return the transcript so now we can return the transcript okay so this is audio to transcript what we will do here I was talking about we don't have to load the model every time if you're running in a in the if you have a runtime environment when you are running it will not have to load every time what we can do we can do hd.cache okay so far good okay we what we did we first have written a function that will load the model you know from whisper and then save audio which will basically uh save the or download the audio from YouTube URL and then you know save uh save it in your directory and then we'll take that audio file and then we'll pass it to this transcript audio transcript function where we are using whisper okay so now let's do one thing let's write the streamlined code where we will take where I have an input box and in that input box and YouTube url will be given and then we perform this operations okay with the help of the function that we have created okay so the first is HD dot markdown let's call it I'm calling it news article generator app from video okay foreign YouTube video URL okay input the YouTube video URL that we are HD dot header and then we will the URL link a variable where you know we have this text input instead of text input and in that text input will have a label that will call enter the video URL here something like that okay let's run this and see if we are getting something okay so what I'll do how do we run htmlit application we just say you can see my virtual environment has been activated extremely um run app.pi so you can see I have news article generator app from video a very minimalistic UI just to test uh your AIO machine learning workflow okay with the help of streamlit okay and I we have something called input the YouTube video we have an input box here now let's do one thing okay what we will do we will write uh let's try it out so what I'll do is say if St dot will have a check box and I will tell you that why we are having a check box here so we'll have a check box which will say start analysis because we have to handle the state I'll tell you what we are meaning by what do you mean by session state in streamlit when you have nested buttons it's very difficult to handle the state you know sometimes so what we'll do we'll have a check box here and we'll have the button for uh download uh the zip file okay within this checkbox okay so be able to handle the state in a better way so what I will do we'll have HD dot check box start analysis and then the first thing let's define two variables so video title we are getting this uh YT dot title right on on Top If You See white hero title which will be written and then we also have to get this audio file name so audio file name and I will use that method that we created save audio so save audio and here Will pass the URL link not the URL so sorry URL link where we are you know have this variable under streamlit okay from the text input so this is the first function that we have written now what we'll do let's see if we are able to get the audio so HD dot audio and then we only have to do audio file name here okay so audio file name Will first try try this out if you're able to get the audio file let's see that I'll do a rerun you can see now there have been some changes start analysis and enter the video URL I will go to this I'll take this video URL the video that we you know we have also seen earlier and I'll just say start analysis perfect so you can see I got an audio file I can play this audio the people you can see the same audio that we have done use this video right the Trump you know somebody is taking interview of trump and he's talking about uh some internal issues going on in the United States earlier you can see the audio file has also been downloaded okay does Donald Trump think you know something okay so let me what do let me do one thing I'll go to this new aggregator folder you can see this is the audio file which has been downloaded does Donald Trump think Muslims are a problem see in an interview with Don Lemon only the file name with extension okay so this is what we have done now so so far so good so we have this audio now okay so let's also get the transcript okay so what we will do we'll say hey there is a variable let's define it transcript and in the transcript we'll say use this function audio to transcript that we've created and just pass the audio file name excuse me sorry audio file audio file name transcript and now what we we can do something you know we can HD dot header or something where you say transcript uh are being generated or you know are being extracted and just you can also use some success message or some status messages like info warnings and whatnot right St dot header has been done now we'll do St dot success I will use a status message and here I will pass my transcript now let's see if it's working okay so what we have done we have this audio file name here and we will take this audio file name and we'll pass to this audio to transcript function that we have written over here what we are doing here is we have you know loaded the model The Whisper based model and we are using that base model here we are taking this audio file which we have saved in the directory and we are generating the transcript basically okay we are running the model and we are extracting transcript and we are returning a transcript over here we will get the same transcript here for the particular audio so let's see that so what I will do I'll go back to my uh excuse me I'll go back to this streamlined application and what we'll do I'll say start analysis again you can see it's still running just pay the audio perfect so if you don't know about uh whisper much what you can do you can save open AI with store and you can come over here you can go to their GitHub repository and you can read the documentation here they have a very neat and clean documentation I was talking about these five different models you can read it here they have different models with different trainable parameters and the relative speed and all these details that will help you and this is the documentation you can use so let's come here you can see transcripts our transcripts are being extracted and here we got the transcript he says Muslims are a problem okay that's what he said are there problem you tell me and something is talking about there okay and I know many are great people and let's do something you know exactly this is what we got right here Trump is trying to you know uh talk good about Muslims and I think attraction is a good thing okay so we have the transcripts here you can see transcripts are being generated now let's go back to vs code and you can see we have used both of the functions so we have used this function to get the audio we have used this function to get the transcripts and now once we have the transcript now we have to do the prompting learning with this transcript right so we have to pass this transcripts to open AI DaVinci 3 Model okay which is the most capable model which is also been flagged as GPT 3.5 okay so let's use this to generate the news article okay how are we going to do that okay so what we have to do first now we have to create a function here let's create a function so I'll create this function I'll say hey Define and I'll say text to news article takes to excuse me text to new article text to news article and here what we have to you know we have to give a text here okay so that will be a text so uh text me and here we write all the code okay from open AI so what we are going to do first I'm going to Define a variable response and the same thing if you if you don't have much understanding about you know this open AI DaVinci three models please watch my previous videos I have covered them in redmi generator video and also the mom generation generator video you can watch that what you mean by response what is the prompt engineering what is the temperature and all other details okay so open AI dot completion open air Dot completion.create and in this create now we'll write all our codes first thing is the model we which model do you want to be use we want to use the Darwin C3 model so this is what we are going to do we are going to do text DaVinci and 003 this is the model excuse me let me just do one thing here will be comma a model and we have the model we have to do the prompt here so what I'm going to do now I'm going to say prompt and this is very important okay so in prompt we have to uh let's do this write a news article you know write the news article in 500 words not more than that and you can do this prompt Engineering in other way it all depends what kind of response you are looking from you know open AI based on your text data that you had okay so 500 words from the from the from the below text and this is the line breaker that I am using and here I will pass my text which has been you know passed here as a parameter so text and this is basically nothing but your transcript the transcript that we got it using whisper you know from that video the audio that we have saved here in the directory right so now let's define the temperature so what I will do temperature which is going to be 0.7 basically it's for Randomness and creativity you can read about it and if you don't have much understanding please watch my previous video to understand about temperature or maximum tokens and penalties right so maximum two ends I want to say hey let's keep it as 600 because we need the word not more than 500 there right so I'm just keeping the max tokens as that and then we'll have top P equal equal to one give me top equal to 1 and then what else we have we have frequency penalty excuse me frequency nit skip this you know zero and presence penalty as well distance penalty has zero that's it so this is the function that we have now what we have to do we have to return this so what I will do I'll say return again the response the variable and we do not need everything what we need response we need only the choices and that we need the first one and then we need only the text so text excuse me text this is what so we have a transcript basically the text parameter here and then we'll be using open a DOT completion.create and then we will we are using this DaVinci 3 Model we are doing a prompt here and then we have temperature and then we have maximum tokens top P frequency penalty and presence penalty that's it so now let's do one thing here guys so what we'll do now so we have SD dot success transcript let's also HD dot header and here we'll call it a news article now here we'll use this result what I'll do result equal to St dot sorry or HD dot now we have to use this function text to news article and here we'll pass the transcript so let's excuse me transcript here we have passed the transcript and let's do hd.success to get the result okay this is what we need okay we had the audio saved here and we have used that audio to get the transcript and now we are passing the transcript to get the news article now let's see if we are able to get the desired response I will again do a rerun let's remove this okay and now let's click on start analysis he said Muslims are a problem okay that's what he said okay are there a problem can you tell me and I say I don't see the people and I knocked down the World Trade Center going back to Sweden okay so he said Muslims are about that was part of what he said well I mean can you make the case that it's been taught by the way you can see transcripts are being extracted here and you can see the news article has been generated here guys this is superb so we have we have this transcript and now we got the news article you know uh within 500 words that's fantastic so you can read and read this let's read this so Donald Trump has made a controversial statement regarding Muslims claiming they are a problem and his comments of spark debate among social media users that have been met with criticism criticism from many in a recent interview with the news Outlet Trump said Trump went on to say that he has great friendship with many people who are Muslims living in his building and a lot of other things that he's talking about it and he he wants to build some kind of policies if you read here you know for them so fantastic so now we have the news article we have the transcript but to use this you have to copy paste and that's we do not want to you know that has happened okay what we have to do we have to help end user so they can download it okay they have to download this news article and the transcript how can they do it okay so we need to keep a button over here which will down help the end user to download in the zip file that's what we want to do let's do that so let's let me go back to my vs code foreign okay so you can see this is uh we have got the result and now we have to write let me just comment it over here so uh write the code to save and download in a gif file or something okay and this is what we are going to do guys okay so write the code of excuse me there's a wrong here okay so that will be right the code so let's write so the first thing is we have to need couple of transcript underscore text a couple of variables here what I will do I'll say open and in this I'll write transcript this is my file name that I'm going to save transcript text W that's very good you have to write this so transcript.take and then we have to use transcript text Dot write transcript and then we'll close this so transcript text Dot close this is what we have to do the same thing we have to do it for article guys so let's copy this and just replace this excuse me and transcript I'll say article and now let's replace this article.txt is my file name and here it will not be a transcript whether it will be uh result the result where we are storing that invariable you see this result we have right where we are getting this new text to news article and then we save this one we so what we have done we have created the file name we have writing it inside that with the help of the variables that we have and now what we have to do we have to write this uh let's call it a variable and we have to use that gif file that we have you know imported the file and will not put transcript.txt you have to bind both in one so what we'll call output dot j and then again it will be right output.zip and then we have to zip file and both of them so first we have to do write G for file right and the first one is transcript Dot txt first one is transcript.txt and the other one is G file Dot uh excuse me right and the other one is uh article so article Dot txt G file dot write an article dot right and then we'll close this so we'll say zip file Dot close so we have what we have done we have bind them together out in in this output.g you know we have these two uh different text file there you want to bind them together in a gif file and then we will have end either download that how we can do that so let's do that so with open and it will not be transcript.txt which has been suggested to me it will be your output.g output.zip and that will be excuse me we have to you know as zip as gift download and now let me use this so let's create a button there so if button HD dot download button excuse me actually download button and here we can write all our labels the first is the label what we want to name that button so I'll say hey download zip this is done we have the label now what next we need the data this is your data which is my nothing but the Jeep download which I have you know here as Jeep download okay so data is done now file name file name we know output.j output dot zip and the MIM okay so application GPS this is correct it will be application perfect so this is what we have to do guys Okay so we have you know created files transcript and article for both of them we have created one more file which is a ZIP file output dot zip and we have bind this couple of text inside that zip file and then we have created this button okay streamlined button where we are writing the labels and the zip downloads which is the file name the data and then we are giving the file name and mime okay so basically now let's run this one so I think we have written all our code here so let me just give you a walkthrough what we did in this we have imported all the required libraries and then we had this EnV file open AI underscore API key which is in this EnV file where I have stored my open AI API you can also get it Go and create an account and get the API key from open Ai and then we have written the load model function where we are loading the whisper based model and we are also doing the SD dot cache which is a decorator where which basically handles to load the model from the cache okay every time you don't have to download the model if you are in the runtime environment and then you have to save audio where what we are doing is we are taking a YouTube video URL and we are only doing only audio equals to True which will only give us the audio file we are hand we are doing some file handling here with the MP3 and path and then we have returned the audio file name and the title once we have the file audio file we are passing it to whisper model and we are getting the transcript out of it now once we have the transcript we are passing it to this open AI WC3 model to get the news article within 500 Words and we are also doing some uh other processing here like temperature and Max tokens and once we got the news article here you can see we have integrated into a streamlit application we have just use the function to obviously represent them on a UI and once we got those things we have created couple of files and bind them in a zip file and the user will be able to download let's do that let's run this one so now we have this video we are using it and now let's run this and see if we are able to get the desired response he said Muslims are a problem okay that's what he said okay are there a problem can you tell me so you can see here we have got the audio here you can see transcripts are being extracted as well in the transcripts word error rate are also very good I cannot see any problem there it looks very fine and the news article has been also generated see this is superb right the news article you can also see the header Donald Trump's controversial comments on Muslims and you can use this completely now what you guys can do here you know you can also extend this application you create us you also add a summary function you add a word cloud function create a good application you know you can create a SAS product out of it it all depends on you how you want to take this forward we have this news article you can also have on option there okay where end user can write their own thoughts they can mix their thoughts into this uh generated article and then they can do paraphrasing or they can you know create a powerful news article out of it all these functions can be added into it let me know your thoughts in the comment box guys if you you know if you have done something similar and you have extended these two for some other purpose right so now we have you can see we have a download zip option here where you can download the G5 now let's see if you are able to download the zip file guys you can see it's asking us to download it so what I'll do I'll say okay let's keep it in the download itself and I have this output.jib file I'll go here what I'll do now okay let me do one thing let me go to okay here I have output.1 you can see I have the transcript txt excuse me let me go to download you can see here we have the output1.j right we'll just take extract here or something you can see output one not one okay you have see both the files article.txt and transcript.txt let's open it to see if we have able to you know write the content there perfect so this is a transcript that we have you can see the same transcript and this is the article that we have see the article that we have perfect isn't it so now you can also write your own you can edit this file you know you can also put some images over here when we are creating an article and you can see all my uh code and responses are here now let me tell you a reason okay why I didn't go along with the button here okay so see if we have a nested button which is the download zip button which is the nested button inside this button if this would have been a button here a streamlined button it's very difficult to handle the stimulate State the State and State Management that we do because estimate I don't think they still provide it you know it's difficult to handle a state within a button or next button that's what what we did we had a check box here stat analysis and then we had a button inside it so it should not be the page should not get refreshed you understood what I'm saying right so if I had a button here and this when you hit this button all your responses will be refreshed and you will not be able to get this response there estimate have created something in react state or something that they are doing it but I don't want that to be happen I want in the pure python okay so you see this is the start analysis I created a check box and here we have a nested button even if you hit the download these responses will right be here it will not it will not be refreshed and you will not you will be able to see it here okay so you can also try it I haven't gone through the latest update of streamlit but it's very important to handle this state okay so in this video guys what we did we you can see that you we in and we build an application where you have an input box where you can input your video URL we we can listen to the audio the knock down the World Trade Center we can listen to the audio we use Pi 2 for this we use whisper for the transcripts and we use open AI GPT 3.5 for this news article and then we you know uh had this option where user can download this zip file now if you want to get this uh all the code you can get it from this AI anytime GitHub repository go and see the news article generator using chat GPT and Whisper you can download it right so guys if you like this video you know uh please like the video and share with your friends and also subscribe the channel and if you have any thoughts and feedback and if you have done something you know similar in part or if you want to do something you know and extend this please let me know your thoughts in the comment box okay thank you so much for watching the video see you in the next video guys