Speech Recognition In Java | Convert Speech To Text

Key Takeaways

In this video, learn to transcribe audio files in Java using Assembly AI's Java SDK. Assembly AI is building the best API platform for developers to build speech AI applications and services for the world to use. With Assembly AI, you can generate speaker labels, hide sensitive information in audio files, and much more. You can also make use of large language models with Assembly AI's lemur endpoint. And that means you can build speech AI applications like these. On top of all of that, with Assembly AI's latest Universal One model that was trained on 12.5 million hours of multilingual data, you can also now confidently transcribe multilingual audio files. Before we get started with writing our code, make sure to check out Assembly AI's documentation for detailed code examples in Java for making use of real-time speech to text transcription uh lemur which is our LLM endpoint as well as a bunch of audio intelligence features. This documentation will also contain the code that we'll be using in this example. So let's get started. I've went ahead and created an empty Java project and also I have installed Assembly AI's Java SDK. Once you've done this, you can go into the application and start writing our code. Now, we're going to be importing a few libraries, namely Assembly AI and Transcript Types. Once we've done that, let's first create an assembly client object. So, this right here is where you'll be writing your API key. And this is what I'm going to be pasting right here. To get a free API key, all you have to do is go on to Assembly's website and create an account for free. Once you've done that, let's move on to the next step where we'll define our URL of the file that we want to transcribe. Next, let's create a transcript object which will contain the transcript from assembly API. Transcript equals to client dot transcripts dot transcribe and then here we want to pass the URL that we've just created. Once we've done this, we want to write a statement which will print out the transcript. Let's save this and hit run. Once we've hit run, as you can see, this is the transcript from Assembly AI's API, which contains a transcript of the file that we've just passed. Next, what we're going to do is retrieve Spa labels from the same transcript. So, we have to make a few changes. First off, let's introduce a config parameter. So right after URL let's create something called params equals to transcript optional params dot builder and speaker labels equals to true do. Once we have this variable, what we want to do is go into our transcribe method and write params. Once we've written that, we should also be printing out each individual speaker label. So what we'll do is transcript getes dot if present So this will enable us to print out each speaker's label as well as what they're saying. So let's save this and hit run. And this is what your output would look like with a speaker label as well as what they have said. As of now, we've been using AssemblyAI's best model, which has the highest accuracy. However, AssemblyAI also offers a nano model, which is a much more lightweight and affordable model, which is also really great if you're doing a lot of transcriptions at once. So, let's check out how you can toggle the nano model. All you have to do is select a speech model by typing in speech model and you can select nano. And that's a super simple way to get started with nano. Next up, check out this video above on how to build a retrieval augmented generation application for multi- speakeraker data.