https://github.com/rickcau/api-transcript-service

https://github.com/rickcau/api-transcript-service

aihelper azure-functions blob-storage meeting-transcription semantic-kernel sk transcript

Last synced: 10 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/rickcau/api-transcript-service
• Owner: Rickcau
• License: mit
• Created: 2024-01-23T00:14:17.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2024-01-31T16:10:09.000Z (about 2 years ago)
• Last Synced: 2024-12-26T04:30:22.367Z (about 1 year ago)
• Topics: aihelper, azure-functions, blob-storage, meeting-transcription, semantic-kernel, sk, transcript
• Language: C#
• Homepage:
• Size: 279 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.txt

Awesome Lists containing this project

README

          
# Updates

1/25/24 - Added support for language translation.

# Transcript Summary API - Blob Trigger

In this example, I provide an example of how leverage an Isolated Azure Function Blob Trigger that parses a Teams Meeting Transcript into a condensed format that is then used by AI to summarize the condensed transcript.  I also expose an HTTP Trigger for Language Translation purposes.  In the HTTP Post request you can pass a FileName, Container and target language and the API will translate the content into the target language.  This example is leveraging the Semantic Kernel SDK so extending it's capibilities is super simple.

# Language Translation API - Http Trigger

When calling the Translation API you need to pass the following JSON structure in the request body.

    ~~~

         {

            "fileName" : "Summary#1baa3e6e-46d7-4039-9581-871766800237#someemail@test.com#.txt",

            "containerName" : "summary",

            "targetLanguage" : "Spanish"

         }

    ~~~

**Important Note:**  I am not using the Azure AI Translation Service for the Language translation, I am simply using the GPT Language translation capabilities.  You could leverage the AI Translation Service by using Semantic Kernel Plugins with a Native Function that calls the AI Translation services.  Maybe I will implement that feature at a later date in this example. 

The **fileName** is a valid file name that exists in the target container.  The target contain needs to exist in the storage account associated with the **BlobConnection** environment variable.

The purpose for condensing the transcript is to reduce the number of AI tokens needed to summarize the meeting.  Simply put, there are better formats that can be used when asking AI to summarize the meeting which makes it more efficient for AI to summarize and reduces token usage.  If needed you could augment the TranscriptService class to chunk the data, especially if the file is large. 

The API will open the file and read it's contents into a memory stream and pass it's contents to a Semantic Kernel Translation Plugin for translation and the translated text will be returned to the client.

## Technologies

1. Isolated Azure Functions with a Blob Trigger

2. Azure BlobServiceClient for writing to containers

3. Semantic Kernel for AI Summarization purposes using Dependency Injection for the Kernel

4. Semantic Kernel SKPrompts and Plugins 

4. StreamReader and Regex for file parsing

5. Environment.GetEnvironmentVariable to read settings from **local.settings.json** and Azure Configuration Settings

## Requirements for this example

1. You simply need to create 3 containers in a storage account; **input, output and summary**

 

2. Rename the **local.settings.json.bak** file to **local.settings.json**, then  set the following values in **local.settings.json** file

   ~~~

        "BlobConnection": "",

        "ApiDeploymentName": "",

        "ApiEndpoint": "",

        "ApiKey": "