How to Transcribe Audio Files with Python

hey and welcome in this project we are going to learn how to do speech recognition in python it's going to be very simple what we're going to do is to take the audio file that we recorded in the previous project and turn it into a text file let me show you how the project works so here is the audio file that we recorded in the previous project hi i'm patrick this is a test123 and if we run our script we get the text transcription of this audio file so like this here hi i'm patrick this is a test one two three so let's learn how to implement this in python so for this project we are mainly going to need two things assembly ai's api to do the speech recognition and the request library from python to talk to assembly ai's api so let's first go ahead and get a api token from assembly ai it's very simple you just need to go to assemblyai. com and create a free account once you have an account you can sign in and just copy your api key just by clicking here and right away i'm going to create a configure file and put my api key here once i've done that now i have a way of authenticating who i am with assembly ais api and now we can start setting up how to upload transcribe and get the transcription from assembly as api the next thing that i want to do is to have a main file that is going to have all my code what i need to do in there is to import the requests library so that i can talk to the assembly ai api so this project is going to have four steps the first one is to upload the file that we have locally to assembly ai second one is to start the transcription third one is keep polling assembly ai's api to see when the transcription is done and lastly we're going to save this transcript so uploading is actually quite simple if we go to the documentation of assembly ai we will see here uploading local file

files for transcription so i can just copy and paste this and change the code as we need it so basically yeah okay we are importing the request library already the file name we're going to get from the terminal so i will set that later just a couple of things that we need to pay attention here basically there is a way to read the audio file from our file system and then we need to set up a headers the setters are used for authentication so we can actually already set this because this is not going to be your api token we set it to be api key assembly ai right and we need to import it here of course all right that's done so we also have a upload endpoint for assembly and this one is api. assemblyai. com v2 upload but you know this might be something that we need also later so i'm just going to put this to a separate value variable and then just call this one here so when we're doing uh when you're when we're uploading a file to assembly ai we are doing a post request in this post request you need to send this post request to the upload endpoint you need to include your api key included in the headers and of course you need the data so the file that you read and we are reading the data through the read file function in chunks because assembly ai requires it to be in chunks and in chunk sizes of five megabytes basically this is the number of bytes that are in there while we're at it we

can already get the file name from the terminal too right so for that i just need to import and inside system the second or the first not the zeroth variable or the argument is going to be the file name and here let's clean up a little bit all right now we should be able to just run a command on the terminal include the name of the file that we want to upload and it will be uploaded to assembly ai and we will also let's print the response that we get from assembly ai to see what kind of response we get again this is the file that we are working with hi i'm patrick this is a test one two three and what we need to do right now is to run python main. pi and the name of the file in this case output. 12. all right so we uploaded our file to assemble the ai successfully and the response what we get is the upload url so where your data where your audio file lives right now and using this we can start the transcription so for the transcription let's again cheat by getting the code from the docs here is the data the code that we need starting from here so this is a transcription endpoint you can see that it ends differently than the upload endpoint this one ends with upload transcript i will call this the transcript endpoint um headers we already have a header we don't really need this anymore the endpoint is transcript endpoint uh json is the data that we are sending to uh or want assembly ai to transcribe so uh we are going to need to give it the audio url we already have the auto url right so um we got the response but we did not extract it from the response so let's do that odo url is response. json and it was called upload url so we're going to give this audio url to here because that was just an example okay and this way we will have started a transcription and let's do this and see what the result is i will run this again same thing all right so we got a much longer response in this response what we have a bunch of information about the transcription that we've just started so you do not get the transcript itself immediately because depending on the length of your audio it might take a minute or two right so um what we get instead is the id of this transcription job so by using this id from now on we can ask assembly ai hey here is the id of my job this transcription job that i submitted to you is it ready or not and if it's not ready it will tell you it's not ready yet it's still processing if it's ready it will tell you hey it's completed and here is your transcript so that's why the next thing that we want to build is the polling we're going to keep we're going to write the code that will keep polling assembly ai to tell us if the transcription is ready or not but before we go further let me first clean up this code a little bit so that you know everything is nicely packed and functions we can use them pre-use them again if we need to so this one is the upload function yes and what it needs to return is the audio url we do not need to print the response anymore we've already seen what it looks like and we need to put the header separately because we want both upload and transcribe and basically everything else to be able to reach this variable called headers for transcription again i will create a function called transcribe and what i need to return from the transcription function is the id so i will just say job id and that would be response. json and id again we don't need this anymore i'll just call this transcript response to make it clear this will be upload response let's call this transcript request so everything is nice and clean this is this and this goes here and for upload response we use it here and we need to return job id all right so now we have them nicely wrapped up in different functions and everything else looks good let's run this again to see that it works oh of course i'm not calling the function so let me call the functions and then run it upload and transcribe but of course i also need to pass the file name to the upload function so let's do that too audio url is not defined audio of course then i also need to pass audio url audio url to transcribe good thing we tried so this will be returned from the upload function and then we will pass it to the transcribe function and as a result we will get job id and then i can print job id to see that it worked let's see yes i do get a jaw body so okay things are working the next thing that we want

to do is to set up the polling function so the first thing we need to do for that is to create a polling endpoint

pulling endpoint so as you know we have the transcript endpoint and the upload endpoint here that's how we communicate with assembly ai's api with polling endpoint it's going to be specific to the transcription job that you just submitted so to create that all you need to do is to combine transcript endpoint with a slash in between and add the job id but the jaw body is a bit awake so i'll just going to call this transcript id so by doing that now you have a url that you can ask to assembly ai with which you can ask assembly ai if your job is done already or not and again we're going to send a request to assembly ai this time it's going to be a get request well i'll just copy this so that it's easy uh instead of post it's going to be a get request we're going to use the polling endpoint instead of the transcript endpoint and we just need the headers for this we do not because we are not sending any information to assembly ai we're just asking for information if you're familiar with requests normally this might be very simple for you but all you need to know about this is that when you're sending data to an api you use the post request type and if you're only getting some information as the name suggests you use the get request type so the result the resulting or the response that we get is going to be called polling response um let's see it's not job id i called transcript id so that it works then we get the polling response and i can also show you what the polling response looks like looks good okay let's run this all right so the it we got response 200 that means things are going well but actually what i need is a json response

so let's see that again yes this is more like it so again we get the id of the response a language model that is being used and some other bunch of information but what we need here is the status so let's see where that is oh yeah there it is so we have status

processing um this means that the transcription is still being prepared so we need to wait a little bit more and we need to ask assembly ai again soon to see if the transcription is done or not what we normally do is to wait 30 seconds or maybe 60 seconds depending on the length of your transcription or length of your audio file and then when it's done it will give us status completed so let's write the bit where we ask assembly ai repetitively if the transcription is done or not so for that we can just create a very simple while loop while true we do the polling and if polling response dot json status equals to completed we return the polling response but if polling response status is error because it is possible that it might error out then we will return um i'll just wrap this into a function i can call this gets transcription results url and while we're at it we might as well also wrap the polling into a function do we need to pass anything to it yes the transcript id we need to pass a transcript id to it um and instead of printing the response we will just return the response so instead of doing the request here all we would need to do is to call this function with the transcript id uh we can pass the transcript id here or might as well i will just call the transcription or transcribe function in here and the resulting thing would be the transcript id from the transcription function and then i'm going to pass this transcript id to the polling function that is going to return to me the polling response uh i will call this polling response data and inside this data so this is not needed anymore um yeah this so the polling response. json is what is being passed i call that the data so i change this to data here and also data here um yeah then i'll just pass the data uh if it's error i can still pass the data just to see the response and what kind of error that we got and here then i'm just saying none all right let's do a little cleanup so we have a nice upload function a transcribe function what we did before was we were calling the upload function getting the audio url and then passing it to transcribe but i'm running transcribe here so i do not need this anymore i still need to pass the audio url to transcribe so then i would need to pass it to here so instead of this i just need to call this function with the audi url um yeah let's put these here actually to make it a bit more understandable maybe instead of passing the string error i can just pass whatever error that was that happened in my transcription then you know we'll be able to see what went wrong uh all right so what we get as a result from get transcription result id is the data and if there is any the error so then let's why not run this and see what the data is going to look like all right so we get something really big let's see maybe i'll just clear this and run it again just so that you know we can see it more clearly all right so we get the id again a language model that is being used etc now we want the results yes it is under text uh hi i'm patrick this is a test one two three is what we get and we also get the breakdown of words uh when each word started and then each word ended in milliseconds confidence of this classification and much more information what we want to do though even though we have all this information we want to write this transcript that is generated by assembly ai into a text file so in this next step that's what we're going to do all right let's come up with a file name for this file uh we can call it actually we can just call it the same thing as the file name plus txt so the file name okay we were using the argument or variable file name too so maybe let's find something else i will just call this text file name and it will be the file name plus dot txt we can also just you know remove the dots valve or dot mp4 or whatever but well let's not deal with that for now so once i have this i will just open it oops i will open it in writing format and inside i will write data text because that's where we have the text information on the transcript if you remember here this was a response we got and text includes the transcription and i can just prompt the user saying that transcription is saved transcription saved we're happy of course there is a possibility that our transcription errored out so you want to cover that too if you remember we return data and error what we can do is you can say if data is returned this happens but if it errored out i will just print there no it didn't work out and the air itself so that we see you know what went wrong okay let's do a little cleanup again i want to wrap this all up in a function we can call the save transcript data and error will be returned from get transcript url it needs the audio url so i will just need to pass all the url here and with that we're actually more or less ready so let's run this and see if we get what we need the transcript saved in a file for that after the after calling the upload function i can move this one here and calling the upload function here i call the upload function and then i call the save transcript function and let's quickly follow that up um i call the save transcript function it calls get transcription result url calls transcribe is here uh it starts a transcription process and then a get transcription result url also calls polling so it keeps pulling assembly ai and when it's done it returns something and then we deal with it in the same transcript function and we either save a transcript or if there's an error we display the error so uh let's run this and see if we get any errors transcription saved all right let's see um output valve. txt if i open it up it looks quite small maybe i can if i open it like this yes hi i'm patrick this is a test one two three is the result that we're getting so that's awesome we actually achieved what we wanted to do so in this next couple of minutes i actually want to clean up the code once again because you're going to build a couple more projects and we want to have a python file that has some reusable code so we don't have to reinvent the wheel all the time so let me first go here actually when we're doing the pulling if we just have the while true loop it's going to keep asking assembly ai for results and you know that might be unnecessary so what we can do is to include some

waiting times in between so it can ask if it's not completed yet it can wait let's say 30 seconds to ask again so we can inform the user waiting 30 seconds what i need is a time module so this call is 30 and i will just import time here and this way it should be waiting 30 seconds in between asking assembly ai if the transcript is ready or not and okay let's create that extra file that we have api communication i'll call it um yes so i will move all of the functions that communicate with the api there so i need to move the upload function transcribe poll all of these actually so just remove that yeah uh let's see did we miss anything no i'll just remove these from here file name can stay here of course headers and the upload and transcript endpoints need to live here because they are needed by the functions in here we have to import the requests library so we don't need it anymore here we need to import the assembly ai api key system needs to stay here time needs to go there and we also need to import from api communication uh import let's just say all and that way we can use these function in our main python script i will run this again to make sure that it is still working so i will delete the text file that was created i will keep the output nice so we also get the prompt that the program is waiting 30 seconds before asking again oh yeah we passed the file name but of course it might not exist there so let's go and fix that the file name is here we only pass it to the upload function and the upload function is here now um and in the save transcript we do not pass it but we are actually using it so what we can do is to just also pass the file name here and that should be fine it should fix the problem transcription saved all right let's see i'll put 12 txc hi um like this hi i'm patrick this is a test one two three so this is a very short audio file and we've actually been using it over and over again so i want to also show that this code is working to you using another audio file this is the audio of the one of the latest short videos that i made for our youtube channel i was just talking about what natural language processing is so this time maybe if i add underscores it will be easier to call yes i'll just copy its name and when i'm calling the script i will use its name this will probably take a little bit longer because the audio file have you been using is only a couple of seconds and this one is one minute so we will see what the results are going to show us right here we go the transcription is saved we find it here right this is exactly what i was talking about let's uh listen to it while the transcription is open can alexa be your best buddy well not now but probably very soon we have been seeing gigantic leaps over the last couple of years in terms of how computers can understand and use natural language all right you get the idea so our code works this is amazing i hope you've been able to follow along if you want to have the code don't forget that you can go get it and the github repository we prepared for you using the link in the description