Use GenAI for languages not supported by Amazon Transcribe Call Analytics

[dumip]

Amazon Transcribe Call Analytics currently supports a limited set of languages, creating a gap for many European languages including Romanian, Croatian, Czech, Dutch, etc. (https://docs.aws.amazon.com/transcribe/latest/dg/supported-languages.html)

I propose to create an alternative analytics pipeline for non-supported languages:

1. Use basic Amazon Transcribe for speech-to-text conversion
2. Implement GenAI-based analysis on the transcribed text to provide similar capabilities to Call Analytics
3. Maintain compatibility with existing Call Analytics structure

Would appreciate feedback on this.

[rstrahan]

Interesting.. I like it. Can you elaborate on which features you would implement in (2) that aren't already part of PCA itself (eg when not using Transcribe Call Analytics, PCA already uses Amazon Comprehend for sentiment, and Bedrock for flexible call transcript summarization).. What other feature gaps would you address?

[dumip]

Here's what I found during my tests with partial supported languages:

1. Sentiment trend is not calculated (because ro-RO not being supported in Transcribe Call Analytics nor Comprehend).
2. PII masking is also not supported
3. Entity extraction which is currently only performed via Comprehend (at least this is my current understanding)
4. Category assignment - which currently seems to be performed only via Transcribe Call Analytics

The Basic Transcribe(and Transcribe GUI widget), Bedrock flexible call transcript summarization and Generative AI Query work as expected.

[bryost]

Hi Dumitro, I like it also. I think even for languages supported by Comprehend, an LLM will do a better job at sentiment analysis, masking, entity extraction. Would like to understand more about the category assignment feature you noted.