Case Study: Amazon Connect achieves real-time multilingual transcription and lower QA costs with Gladia

Amazon Connect improves transcription accuracy by 29% with Gladia

Amazon Connect, the AWS contact center service, was struggling with the high costs and limited accuracy of its native transcription tools, particularly for multilingual and accented speech. This impacted downstream processes like CRM population and quality assurance. To overcome these limitations, Amazon Connect integrated with Gladia, utilizing their Solaria-1 speech-to-text model for a more robust transcription layer.

Gladia's solution involved routing call audio from Amazon Connect through Kinesis Video Streams or S3 to its API, bypassing Lambda timeout issues and bundling features like PII redaction. The implementation resulted in significantly higher accuracy, with Gladia's Solaria-1 model achieving an average 29% lower Word Error Rate across multiple languages, which improved CRM data integrity and enabled automated QA at scale.

Amazon Connect