Amazon Lex

Amazon Lex is Amazon Web Services' managed conversational AI platform, significantly updated in 2024 with native Amazon Bedrock integration to bring large language model reasoning to production voice agent deployments. It operates as a fully managed cascade: Amazon's proprietary ASR handles speech recognition, an Amazon Bedrock foundation model handles intent and dialog management, and Amazon Polly's 90-plus neural voices handle synthesis. This eliminates the need for developers to assemble and orchestrate separate speech components, making it the lowest-friction path to a production voice agent on AWS infrastructure. Amazon Connect integration gives Lex full SIP telephony support, enabling deployment as an AI IVR replacement for enterprise contact centers. The platform supports 13 language locales across English, Spanish, French, German, Italian, Japanese, Korean, Portuguese, and Mandarin Chinese. AWS Lambda function calling is native, and streaming conversations allow multi-turn voice sessions in a single persistent WebSocket connection. Pricing runs at approximately $0.004 per speech request, translating to around $0.04 per minute at typical conversation cadence when including Amazon Connect voice charges. Amazon Lex is the natural starting point for AWS-native development teams building enterprise voice agents. The familiar IAM permissions model, tight integration with S3, DynamoDB, and Lambda, and direct path to Amazon Connect's SIP network reduce onboarding friction for teams already on AWS. STT accuracy at around 10 percent WER and 800ms latency trail best-in-class specialist providers, so teams with strict accuracy or latency requirements should evaluate Deepgram or OpenAI realtime alternatives. Where Lex wins is operational simplicity and enterprise telephony integration within the AWS ecosystem.