Voice agent on AWS

Estimate a voice bot: Transcribe (STT) + Bedrock + Polly (TTS) + Lambda.

CallerTranscribe (STT)LambdaBedrock modelPolly (TTS)

Example estimate

Use this for phone or voice-channel assistants. Speech-to-text minutes and text-to-speech characters usually rival the model cost, so all three are broken out.

$3,989.11/month

30,000 conversations/month · 4 min each

Estimate your workload

Bedrock model

Model choice is usually the biggest cost lever. Nova Micro/Lite are ~10–70× cheaper than Premier.

Conversations per month

Minutes per conversation

Input tokens per conversation

Output tokens per conversation

TTS characters per conversation

Results update automatically as you type.

Result

$3,989.11/month

30,000 conversations/month · 4 min each

Transcribe (STT)(4 min/convo): $2,880.00
Bedrock (Amazon Nova Pro): $148.80
Polly (TTS)(2,000 chars/convo): $960.00
Lambda (glue): $0.3060
Per month(all services): $3,989.11
Per year: $47,869.27


Transcribe (STT)	$2,880.00	72%
Polly (TTS)	$960.00	24%
Bedrock (Amazon Nova Pro)	$148.80	4%
Lambda (glue)	$0.3060	0%

Scope: covers inference, AgentCore and the managed services this architecture uses. Excludes raw EC2/GPU compute and committed-use discounts (too volatile to keep honest). Region-agnostic planning estimates.

Frequently asked questions

Which cost usually dominates a voice agent?+

It depends on call length. Long calls make Transcribe (per-minute STT) the biggest line; short, chatty calls tilt toward Polly (per-character TTS) and the model.

Other AWS architectures

Amazon Bedrock model pricing

Estimate the token cost of any Amazon Nova model on Bedrock, before you build.

Bedrock AgentCore agent cost

Predict a Bedrock AgentCore agent bill: model tokens + Runtime + Memory + optional tools.

RAG chatbot on Bedrock

Estimate a retrieval-augmented chatbot: OpenSearch vectors + Bedrock generation + Lambda.

Batch document pipeline

Estimate a high-volume document extraction/summarisation pipeline with batch pricing.

Minimal serverless Q&A

The cheapest starting point: API Gateway + Lambda + Bedrock, no vector store.

← All AWS AI cost calculators

AgentCore and managed-service rates last verified 14 July 2026 (us-east-1). Bedrock model prices last reviewed 28 July 2026. Estimates for planning, not a quote.