deepgram

An AI-powered speech recognition platform offering advanced speech-to-text and text-to-speech APIs for building voice-enabled applications.

Deepgram

Hey Everyone and Welcome back to our channel. We post daily videos on cutting ash technologies. In this video we are gonna be creating an amazing talking assistant just like alam's into speech using some platform. For that we are gonna be using Deepgram. 

The best part about it is that it is absolutely free and it provides you the feature to convert your speech into text and text into speech. It also contains sample code for both of the scenarios. So let's head back to the browser. 

Here I am inside the documentation of the Deepgram. In order to get the code for the text to speech feature, you have to go to the navigation menu, find the text to speech, and click on getting started. Once you land on the page, you have to scroll down a little bit to the part which says make the request with the STK. 

So this is the code that we need. You simply have to copy this code from here, head back to your VS code, and create a new file. I'm gonna call it TTS, which is text to speech, and I'm going to paste all the code right here inside this file. 

Here inside the code, I've made a few changes. Firstly, I've enclosed all of the code inside a function called tts, which takes a text string as an input. Then here in the code, you can see that it speaks options. It has hard coded the text, but here in the code I have set the speak option, which is the variable to be passed, and I have set its value with the text which will be passed as a function parameter. 

This text will be converted to speech. Then everything is the same as the original code except for one thing. Here in the original code, you can see that after getting the response, it is printing the response after converting to json. But I want the file name in which the response is saved. 

So here is a variable called file name, and I am simply going to return the file name. Then in order to test it out inside the main, I am simply calling this function with the hello this is a test. If the code is working fine, then the text to speech. 

Feature will actually convert this text into speech by providing as a new file by the name Output Web. So now let's go ahead and run this file.

To see the output and you can see that the project executed without any error. Inside the left navigation menu we have a new file called output web. If I click on it because of some reason it is not loading the web view here. So I am gonna head to the project directly. Here is the output web file. Let's quickly open it and you can see that it has converted our text into speech. So this means that a text to speech feature is working fine.

Ok after a got to tell one thing that in order to use the Deepgram you have to enter the Deepgram API key. So I have created DocumentsFragment account and inside the dashboard after signing in you will land on the page that looks like this. You will get a credit off about $200 and after that your plan is pay as you go. So in order to get the API key simply click on the API keys from the left navigation menu and click on the button to create a new API key. Provide a name, set the permissions and then it will provide you the API key which you have to copy and use it inside the avi file. Then you will be able to use the Deepgram in order to perform text to speech.

Now we need to create the front end of our application and for that I'm gonna create a new file. I'm gonna call it app.py. This file will contain the front end of our application and the application is gonna be the streamlet application. So firstly I'm gonna import streamlet as sd. Then in addition to streamlet we're gonna be needing three other libraries. Firstly we need the button library from the b.m. Visit package. We're gonna be needing the custom Jess and we'll be needing the streamlet bouquet events.

After that I'm gonna define the main function. Inside the main function firstly I'm setting the title. Then I'm writing please click the below button to stop the recording. After that I'm creating a button. The label on the button is gonna be speak and it's actually not the stream lid button but the button from the b.m. 

Wit library. And this is the button which we're going to use to add the custom JavaScript as well as the Streamlet bouquet events. So once you've created to display the button, we're gonna run the bouquet chart function of the streamlet library inside which we're gonna pass the button. 

And this line is going to show this button on our interface. After that, we're adding a JavaScript event on the button. The event is gonna be the button click and what should happen on the button click. For that, we are using the custom JavaScript and for the code, we are actually initializing the web cued screen speech recognition. 

And we are setting the continuous and the interim results to be true. Once the speech is recognized, then on the result, we are calling a function where this actually contains the audio file. Using this for loop, we are going to transcribe everything which the user has input using the microphone. 

Then we are gonna check if the value is not equal to null. This means that it contains some transcription. Then we are using the document dispatch event function inside which we are creating a custom event by the name get text. 

The detail of this function is actually the value that contains the transcription of the speech which the user has recorded. Then we are gonna initiate a streamlet bouquet events inside which you have to pass certain things. First one is gonna be the button on which you want to add the bouquet event. 

It is the stt underscore button. Then the event is the get text event which we have just created. Then the key is to listen. We have set the refresh on update equal to false. The override height is 75 and the debounce time is zero. 

Once ten Streamlets Bokeh event containing the custom event of the get text is executed, then the result of this event which is the transcription of the pitch is executed. Then the transcribe result will be stored inside this variable called result. 

Next, we are gonna test that if the result is not empty, then we will check if get text is present inside.

The result then get text is actually the event. And the custom event which we have created inside the result we are gonna use to get the result.

Function to get the value of get text. I am simply printing it on the screen. Alright, so in the app.py file up till now we are firstly recording the user input using the microphone. Then we are transcribing it by converting it into a transcript. We are finally displaying the transcript. 

So let's quickly go ahead and run all files to see the output. For that, you have to run the command sms. Let's just hit enter and a new browser window will be opened that will show you two buttons. One is for speaking and the other one is for stopping the recording. 

So now if I click on the speak button and say hi this is test, here is the output. You can see that if I again click on the speak button right here and say here is the output, you can see that it is working absolutely fine. Our recorder is working fine and it is perfectly transcribing our speech. 

So now one module of our application is done, which is to record using the microphone. But we need to pass the transcription of this recording to an AI model in order to get the response from it. Then that response will be returned in the form of speech to the user. 

So for this application, we are going to be using the Grok AI. As you already know, Grok AI provides you free credits and it also provides you a large number of models which you can use. In order to use the Grok AI, you need to have the Grok API key. So head to After signing in, you will see on the left navigation menu an option for the API. Simply click on it. 

The Create API Key button and then name for the API key click submit and it will generate a new API key for you and then you simply have to go to your avi file create a variable for the GROC API key and then paste your API key Right There Inside the Avi File. 

Okay So Once This Is Done Now I'm Gonna Head to the Grock Playground and Here You Can See That We Have Different Option from the right hand side you can select any model that you want. I'm gonna go with the l 370 b 8192 model which is the latest one and as you already know that Grock is absolute lightning fast. 

So if I ask it to write a poem On clicks and submit the result you will see that in just a matter of single second it has been created Fools but it's what we're testing actually want to get the code behind all of this conversation and for that I'm gonna click on the view code button and you can see that here we have the inspiration. 

I'm simply going to copy this code from Here and I'm gonna head back to my VS Code. I'm gonna create a new file I'll call it grok a.py because it Will contain the code for the grok a and I'm going to paste the entire code right here. 

Now I'm gonna make a few changes to the code I only want a single message which is my message in order to test it out so I'm gonna Do that. Then I want all of these chunks to be present inside a single variable for that I'm going to create a New variable called response. 

It is going to be an MT string and here I want to concatenate all the different chunks of the response inside this single variable and here I actually want to remove this and part. OK so this for loop is going to hydrate through all The chunks and will concatenate them inside a single variable called Response. 

And then finally I'm gonna print the response variable alright so once that is done simply go ahead and run the file in order to see the response. OK so the file is running OK I place equal to plus instead of plus equal to save it and Now I'm Gonna Run the File One More Time in Order to See the.

Output you will see that a file is running and it has displayed as a poem as we have prompted it that write a poem on Slur. So it has provided as a poem on Slur in just a matter of few seconds. So this means that I grabbed portion of the code is working fine now. 

We only need to integrate a three files together. So firstly I'm gonna enclose all of this inside a new function. I'm gonna call it generate response and this function is going to take a prompt as an input. That prompt will actually go here as the content instead of hard coding the prompts. I'm gonna pass the prompt variable right here. So whatever the prom will actually be passed to the Grok using the modern Llama 3 and it will generate the response for us. 

We are simply returning the response. So after getting the response in the text, we need to convert it into speech. For that I'm gonna head to my add pie file and I'm gonna make a few changes to the code. I'm gonna scroll all the way down and I'm gonna remove the file. 

Here first we want to pass the text of the audio which the user has entered using the microphone to the Grock AI. So I'm gonna call the generate response function and I'm gonna pass the get text which actually contains the nest of our response. But right now it's a given I got an error because this function is present inside a separate file. So now I'm gonna import it i.e. 

Gonna say from Grock a a import star so it is going to import all the functions which are present inside the Grock a.py file. 

Now we get the response from the Grock a.py. We need to convert it into speech using the text to speech file which we've just created my text to speech file. 

I'm gonna make a few changes first.

I'm gonna remove this calling to the function and then I'm gonna head to the add pa file. 

And just like I imported all the function of the grok ai file, I'm gonna do the same for the text to speech file as well. Then below it, I want to pass this response to the text to speech button so that it can convert our text into speech. For that, I'm gonna call the text to speech function inside which I'm gonna fast the response. 

And this function will actually return as the name of the audio file, so I'm gonna store it inside a variable called file. We want the response to be automatically played instead of playing the button. For that, I'm gonna head to the top of the main function and outside the main function, I'm gonna define a new function called autoplay audio. 

Inside this function, as a parameter, I am providing a file path. Then using the open function, I am opening the file which is present inside the file path. I am reading its content, storing it inside a variable called data, and then we are decoding it using the Base64. 

Once we've decoded it, now we're simply passing it to the audio controls auto player that will play our file. So in order to use the Base64, we need to import the Base64. So I'm gonna import the Base64 right here. 

Okay, so this function is going to autoplay the audio and it takes the file and the file path as an input. So now I'm gonna head to the end of my file and once I get the file, I'm simply going to call the autoplay audio function inside which I'm gonna pass the file. 

So now if I go through the flow of this file one more time, you'll see that once the main function is called, all the components of the user interface will be created. DocumentsFragment response function which will interact with the Grock AI using the Llama 3 model and will generate the response. 

Once we have the response in the textual form, we will pass that response as a function parameter to the text to speech function which will generate a file containing the audio for us and then we are simply calling the. 

Autoplay audio function by passing the file inside it and that the entire code for an application. Now let's quickly go ahead and test it out for that I'm gonna open a new terminal and I I'm gonna run the command stream no.

Let's enter and you can see that now the new browser window is open where we have our application. So now let's go ahead and test it out. Firstly, I'm gonna click on the peek button right here in order to start the recording. What's the end? Wait for it for the response and you will see that it is running right here and after a few minutes there is a response. 

Juicy sweet and crunchy bruises grow on apple trees as the Messika else are one of the most widely consumer in the world. A are staple in many reasons. Here are some interesting facts about else and you can see that it judges we provide as a detailed introduction of the apple. 

So now instead of asking it some questions, if I simply go ahead and say hi, let's see how it responds. High, you can see that it is running and let's wait for it to be nice to meet you. Is there something I can do with? Ok, so you can see that it judge generated a response for my single high as well. 

So now if I test it out with another promoter, this time I'm gonna ask it what sound does a cat make. So I'm gonna click on this button and say the prompt what sound does the cat make? A variety of sounds but the most common one is one. And in this way, you can ask it any question that you want and it will provide you the answer within the second using the Grok AI Datchi. 

Note, that there may be slight differences in text and video. Watch video for the original content.

Deepgram Video