Speaker 1: Hello, my name is Peyton, and welcome to the next video in the Azure AI learning series for call center transcription and analytics. Today, we'll walk through how to leverage Azure AI language and speech services for efficient call center analysis, and to get even more insights on customer interactions with call center representatives. First, let's start by listening in to a quick call center conversation with a Contoso Pharmacy.

Speaker 2: Hi, thank you for calling Contoso Pharmacy. Who am I speaking with today?

Speaker 3: Good afternoon. My name is Mary. I'm calling about a refill for my prescribed medications. I have been trying to get a hold of someone for weeks and was told that I would get a call back regarding my situation, but it's been weeks and no one's contacting me, so I thought I'd call.

Speaker 2: I understand your frustration, Mary. Can you tell me what exactly you're trying to accomplish?

Speaker 3: Yes. I'm trying to get a refill of my prescription drugs that my doctor prescribed to me for cholesterol.

Speaker 2: Okay. Certainly happy to check that for you.

Speaker 1: Now, let's say that you want to learn more about this and other similar conversations in your team's call center. First, we can use real-time speech-to-text using the Azure AI speech service to transcribe an audio file into a conversational text file. Here's a glimpse in how to try out this feature for free in the speech studio. Here's a look at the transcript provided by the Azure AI speech service. Notice that the transcript is separated by speaker in order to analyze the conversation at the speaker level. Using this text file from the speech service, let's take a look at the language service to see what features can help us use and analyze this conversation further. The language studio is a great stop for trying out the Azure AI for language service. Let's explore the wide suite of language features to support this conversation scenario. First, because you're working with personally identifiable data at the pharmacy, you can use the personally identifying information or PII extraction feature to automatically identify, categorize, and redact sensitive information in unstructured text including phone numbers, email addresses, names, and other forms of identification. Using this feature, the PII model has identified a number of entities that potentially contain PII data. Because the prediction is also provided in a JSON format, you can customize which outputs matter for your business case. For the purposes of our call center conversation, let's say that we're only interested in redacting names from this transcript. Now that we've processed this data, we can see that the model has identified Mary as a name entity and redacted it to protect Mary's privacy. Next, we can leverage the sentiment analysis feature to track the sentiment of the speakers throughout this conversation. Sentiment analysis can help you find out what people think of your brand or topic by mining text for clues about positive or negative or neutral sentiments. For today's demonstration, let's input this text into the sentiment analysis model in the Language Studio. As you can see, the model produces a visual representation of the sentiment score for each utterance, as well as a JSON file for you to use in code within your implementation. Here's an example. Last, since this is such a long conversation, you can utilize the summarization feature of the language service to get a summary of the conversation. We can choose to generate two types of summaries using this API. Extractive summarization, which provides a summary by extracting important sentences within the conversation and abstractive summarization, which generates a summary that captures the overall main idea of the conversation. For the purposes of our scenario, the abstractive summarization may help to give us more clear and concise summaries for quick use by the call center representative or an escalating manager. From here, you can use the power platform to transform this data from JSON format into a user interface for your call center representatives to use day-to-day. Feel free to see more of this data in action using the call center and transcription analytics tryout experience here on the Language Studio. You can also check out how to leverage these AI model predictions using the REST API, a detailed JSON output, or a data ingestion client tool. See this and other resources about the post-call transcription and analytics capability in the description area. Thank you.