Practical guide
Amazon Connect prompt audio requirements
Amazon Connect plays prompts over the phone network, which carries 8 kHz audio. Getting the file format right avoids failed uploads, silent prompts and muffled audio. The rules depend on where the prompt will live.
Prompt library (uploaded in the console)
- .wav files only.
- Less than 50 MB and less than 5 minutes.
- AWS recommends 8 kHz. Higher sample rates, such as 16 kHz, are down-sampled and can lose quality. Use mono. 16-bit PCM is the safe default.
- There is no bulk upload in the console, API or CLI. Each prompt is created one at a time.
Prompts played from an S3 bucket
- The file must be 8 kHz, mono, U-Law .wav, otherwise AWS says it won't play correctly. Choose S3 prompt in the Prompt Studio, or 8-bit U-Law in the converter.
- The same limits apply: under 50 MB and under 5 minutes.
- S3 prompts work in the Play prompt, Get customer input, Loop prompts and Store customer input blocks.
- The bucket policy must allow
connect.amazonaws.comto uses3:ListBucketands3:GetObject, scoped to your account and instance ARN. The Prompt Studio fills this in for you. - Amazon Connect can't play prompts from a bucket encrypted with an AWS managed key. Use a customer managed KMS key that allows
kms:Decryptforconnect.amazonaws.com. - For opt-in regions such as Africa (Cape Town), the bucket must be in the same region as the instance.
Making prompts sound good on the phone
- Trim the silence at the start and end. Dead air before a greeting makes callers think the line has dropped.
- Even out the loudness so every prompt plays at the same level. The studio targets an average of -20 dBFS with a -1 dBFS peak ceiling.
- Avoid clipping. If the checker reports clipped samples, re-record at a lower input level. Clipping can't be repaired afterwards.
- Listen as a caller would. The telephone preview cuts the audio to the 300–3,400 Hz band of a phone call. Music and deep voices lose the most.
Recorded audio or text-to-speech?
Record prompts that rarely change and need a brand voice, such as greetings and hold messages. Use text-to-speech for anything that changes or includes customer data, such as balances, dates and queue position. You can build those prompts with the SSML Prompt Builder.