Should I use Amazon Polly for an IVR system?

Yes—Amazon Polly can be a good choice for an IVR system, especially if you want natural-sounding voices, easy scaling, and integration with AWS services.

When Polly is a good fit

Use Polly if you need:

• Text-to-speech for prompts like “Press 1 for sales.”
• Scalable, low-ops infrastructure
• Multiple voice/language options
• Integration with Amazon Connect, Lambda, or other AWS services
• SSML support for controlling pronunciation, pauses, and emphasis

Things to watch out for

Polly may be less ideal if:

• You need ultra-low latency with no network dependency
• Your IVR requires very dynamic, real-time prompt generation at huge volume
• You need tight control over telephony audio formats and caching strategy
• You want a fully offline/on-prem IVR with no cloud dependency

Best practices for IVR use

• Pre-generate and cache common prompts instead of synthesizing them every call
• Use SSML for numbers, dates, acronyms, and pauses
• Choose a voice that sounds clear over telephone audio, not just “natural”
• Test in 8 kHz mono telephony output to make sure it sounds good on calls
• Monitor latency and cost if prompts are generated frequently

Bottom line

If your IVR is in the AWS ecosystem and you want a high-quality TTS engine, Polly is a strong option. If your IVR is extremely latency-sensitive or offline, you may want a different architecture.

If you want, I can also help you decide between Polly vs Amazon Connect’s built-in speech features vs other TTS engines for your specific IVR setup.