AWS AI Services Workshop

Stop guessing, start reasoning: Adding ubiquitous intelligence to your applications (ASR + NLP chapter)

Welcome to the workshop, prepare to get your hands dirty!

Workload description

Today, as participants of this workshop, you are all staff members of Octank Inc., a global enterprise, active in all verticals you can think of. For all B2C activities, it’s very important for Octank to offer a broad range of contact channels for their end customers. One fundamental channel is their contact center where customers can dial-in to speak with an agent.

Photo by Pavan Trikutam on Unsplash

In order to enhance the customer experience, Octank would like to analyze these conversations for sentiment and key words, to get an idea about customer satisfaction and predominant concerns. Of course such analyses would only take place in an anonymzed manner and after the customer has agreed that their call is being recorded.

In this workshop, you will help Octank with prototyping a new environment with a cloud-based contact center and a processing pipeline for the analysis of call recordings. Octank follow a cloud-first strategy on AWS and they distribute their IT resources globally, making use of several AWS regions. For the prototype, they want their virtual contact center to be running in the AWS region in Frankfurt (eu-central-1).

However, Octank are also setting up a globally distributed data platform around an Amazon S3-based data lake at the moment, with the first leg being planned to reside in the AWS region in Northern Virginia (us-east-1). The above mentioned call recording analyses are to be performed within the central data platform. That means that call recordings need to be replicated from their data source (the virtual contact center in Frankfurt) to the data platform (in Northern Virginia) for ingestion.

For an overview of the AWS global infrastructure, pay a visit to

Manual soldering

Photo by Nicolas Thomas on Unsplash

In labs 1 - 5, you will launch a cloud-based contact center, create some sample call recordings, and set up cross-region replication of the call recordings. Then you will start exploring the call recording analyses manually in the AWS console, starting with a transcription (speech-to-text) of a call recording. You will then go on a short excursion, looking into translating the transcripts into different languages. Finally you will run sentiment analyses on the transcripts and extract keywords.

Processing pipeline

Photo by Quinten de Graaf on Unsplash

After you did this manually in the AWS console, you will use lab 6 to set up a processing pipeline that does this automatically after a new call is recorded. Along with this, you’ll have a look how we can run SQL-queries on our findings and visualize the results.

Expected outcome

By the end of this workshop, you’ll be able to:

  • Launch a cloud-based contact center.
  • Automatically replicate objects from one AWS region to another.
  • Run speech-to-text jobs in order to get a transcript from an audio file.
  • Translate existing texts from one language into another one.
  • Run NLP analyses on an existing text.
  • Run SQL-queries on text files and visualize the results.