Skip to content
TrustList
News

Twilio will fail over live call transcription to a second speech provider from 22 October

Editorial

By TrustList Editorial

From 22 October, Twilio starts a real-time transcription on its other speech provider, Google or Deepgram, when the first does not connect. It is on by default and can be switched off per sub-account.

About Twilio will fail over live call transcription to a second speech provider from 22 October

Twilio will fail over live call transcription to a second speech provider from 22 October

22 September 2026 — Twilio is changing what its Voice real-time transcription service does when a speech provider is unavailable. From 22 October 2026, if the provider a customer has chosen does not connect at the start of a call, Twilio will start the transcription on its other provider instead of returning an error. The change was posted to Twilio's changelog on 22 September, and it will be on for every existing account unless the customer opts out.

What changes

Twilio's real-time transcription runs on two third-party speech engines, Google and Deepgram. Under the new behaviour a transcription that cannot start on Google starts on Deepgram, and the reverse. Twilio says failover happens only when a session starts, never part-way through a call, and that audio stays within the customer's existing region.

The two providers do not support identical features. Where the backup cannot honour a setting, such as profanity filtering, custom vocabulary hints or a particular language variant, Twilio says it will drop or downgrade that setting and still return a transcript.

New customers will default to an "Auto" mode in which Twilio chooses the provider for each session. Existing customers keep their current provider setting unless they switch to Auto themselves. Failover can be switched off for each sub-account in the Twilio Console. The announcement refers to Twilio's Predictive and Generative AI/ML Features Addendum and says nothing about pricing.

Who is affected

Contact centres, sales-call analysis tools and software suppliers that build Twilio transcription into their own products. It deserves attention in three situations: where your contracts or data protection records name a single speech provider; where your customers have been told which provider processes their call audio; and where you rely on filtering or custom vocabulary, so that a silent downgrade would affect quality or compliance.

What to do

  • Before 22 October, decide for each sub-account whether failover suits you, and opt out where it does not.
  • Check that both providers appear in your records of processing, your sub-processor list and any customer-facing privacy information, and read the addendum Twilio refers to.
  • Test transcripts on the alternate provider for your languages and settings, so you know what a failover transcript looks like before one reaches a customer file.
  • If you buy a product built on Twilio, ask the supplier whether it will accept the default or opt out, and whether its sub-processor list will change.

Not yet independently verified. This rests entirely on Twilio’s own changelog entry. No independent report of it was found, so every date here comes from the vendor's own page and is only as reliable as that page. We will update this when it can be confirmed, and remove this note.

Sources