I am trying to determine if AWS Lex uses AWS Transcribe for prompt confirmations. For example, Lex asks "What's your phone number?", the user responds with "1-2-3-4". Lex then asks, "Did you say 1-2-3-4?". What does Lex use behind the scenes as an ASR to determine the user said "1-2-3-4"? Is it AWS Transcribe or something different?
1 Answers
I don't think anyone knows besides AWS but it does make sense for Lex to use the power behind AWS Transcribe
(and AWS Polly
for returning speech from Lex). Speaking for personal experience, till about a month before Transcribe was announced at reInvent 17, I was using Lex
to perform STT (speech to text). This was then possible as intents
could be ignored & passed on to the Lambda
handler. The JSON packet given to Lambda
handler contained the recognized speech (as text), and I returned that back to the caller. However, sometime after they announced Transcribe
, this stopped working, as in the intents
could no longer be ignored. Any input besides those in intents
would return the configured error response. My guess is they stopped this as they launched Transcribe.
Addendum: AWS Transcribe
is pure ASR (auto speech recognition or speech to text). It returns the recognized speech, and meta-data (confidence etc.).
With AWS Lex
you can design your own bots to auto-respond to queries (like in Alexa)

- 2,440
- 1
- 21
- 37