Project Transcribe
We can best introduce our internship project by its purpose: making your daily work life easier. And how would it do so? Let’s assume you are the average consultant. Then you probably have multiple meetings per week, if not multiple meetings per day. We know how easy it is to get distracted during a meeting when taking notes or writing something down, so we developed a software application that does that for you. With our service you’ll be able to pay full attention to the topics being discussed.
What Transcribe does for you
By leveraging multiple AI services, our application takes an audio recording and transforms that into a summarized briefing which it delivers to you by e-mail. This saves you the time and effort of manually transcribing and distributing meeting notes. It’s our goal to make the user experience as convenient as possible, so we designed our user interface to be intuitive and minimalistic. Recording audio is possible under the “Audio”-tab, and supplying all other necessary data is possible through a simple form under the “Data”-tab.
Once the recording is submitted, the audio is automatically processed, transcribed, and summarized before being sent to the designated recipients. To see this in action, check out the following example of an AI-generated recording:
Application Flow
In our application we make good use of various Amazon Web Services to handle your request.
Our primary tool is AWS Transcribe, converting audio recordings into text transcripts. One of Transcribe’s most powerful features is Automatic Speech Recognition, enabling us to support transcription in over over 100 languages.
Additionally, with AWS Transcribe you can enhance your transcriptions with specialized services at an extra cost. For instance, Transcribe Call Analytics provides metrics on customer service phone calls, and Transcribe Medical is tailored for medical terminology and automatic redaction of sensitive patient data.
To summarize these transcripts we use AWS Bedrock, specifically leveraging the Amazon Titan Text AI-model. Titan is a cost-efficient generative AI model, which excels in text processing. Our interaction with this model is straightforward: we input the transcript into the prompt and request a concise summary.
After we have our full and summarised transcription, we deliver it per e-mail to our user’s inbox using AWS Simple Notification Service.
In conclusion
Despite the project’s limited development time of 2 months, we came across several potential enhancements or features that could be implemented in the future. Such as the following:
- E-mail customization: Customise the content received
- Live transcription: Real-time conversation subtitling
- Integrating with existing chat applications
Looking back on this project, we gained hands-on experience with a variety of Amazon Web Services such as IAM, API Gateway, Lambda, Transcribe, Bedrock, and SNS. We also learned to work with the Java and Node AWS SDKs, and understood the fundamentals of Terraform for deploying these services. Setting up a CI/CD pipeline in accordance with best practices further solidified our understanding of DevOps.
Last but not least, we’d like to thank our mentors Femke Tack and Lander Mariën for their guidance, support and inspiration throughout the assignment.
