Vendor watch

Twilio Adds Provider Failover to Voice API Real-Time Transcriptions

Twilio now handles real-time transcription failover and provider selection natively. What this means for your voice applications and fallback logic.

Published October 24, 2026 5 min read By the Breakwatch team 1 source

Voice applications demand high availability and low latency. When a user speaks into a phone system, the software must transcribe that speech immediately to drive the next interaction. If the underlying speech recognition engine fails to connect, the application usually returns an error, leading to dead air or a dropped call. To prevent this poor user experience, engineering teams historically built complex custom routing layers. They had to maintain active integrations with multiple transcription vendors. They also had to normalize the disparate response payloads from those different vendors into a single format their application could understand.

On October 22, 2026, Twilio removed much of that architectural burden for its users. The company introduced Provider Failover and an Auto mode for Real-Time Transcriptions. The API now handles the provider switching behind the scenes. This update fundamentally shifts the responsibility of multi-provider resiliency from the application code to the API platform itself.

The Mechanics of Provider Failover

Until this release, a Twilio transcription session relied entirely on a single speech provider being reachable. If that specific provider experienced an outage or a network blip at startup, the session failed. Now, Twilio automatically retries an alternate provider. The system routes traffic between Google and Deepgram. If Google fails to connect, the session automatically starts on Deepgram. Conversely, if Deepgram is unreachable, the system falls back to Google.

This failover mechanism is now active by default for all existing Real-Time Transcriptions accounts. Teams no longer need to write custom routing logic or normalize the response payloads themselves. The API manages the transition natively and returns a consistent data structure.

There are specific architectural constraints to this behavior. The failover mechanism only triggers at session startup. If a connection drops mid-session after successfully connecting, the system will not switch providers on the fly. Additionally, the fallback routing strictly respects data residency boundaries. Twilio ensures that the failover stays inside the customer region. A session processed in Europe will only fail over to an alternate provider located in Europe, ensuring compliance with local data protection requirements.

Twilio Transcription Failover Flow. A flowchart showing how Twilio routes a transcription session to a primary provider, and falls back to an alternate provider if the connection fails at startup.

Feature Downgrades and Graceful Degradation

Automatic failover introduces a new variable for developers to manage. Different speech providers offer different capabilities and configuration options. When Twilio switches providers at startup, the alternate provider might not support the exact configuration requested by the application.

In these cases, Twilio prioritizes availability over feature strictness. If the backup provider lacks support for profanity filtering, custom hints, an exact language variant, or provider word timing data, the API simply drops or downgrades that specific setting. Twilio "returns a working transcript instead of an error."

This is a classic engineering trade-off. A degraded transcript is usually better than a dropped call or a failed interaction. However, downstream application code must be prepared for this missing data. Consider a user interface that relies on provider word timing data to synchronize subtitles on a screen or highlight words as they are spoken. If the failover provider does not supply this timing data, the application must not crash when that data is absent from the payload. Developers must write defensive code to handle missing fields gracefully. Once the primary provider recovers, the original settings apply as normal on the next session with the usual provider.

Dynamic Routing with Auto Mode

Alongside the failover update, Twilio launched a new Auto mode. This feature changes how developers specify a transcription engine for their voice applications. Instead of hardcoding a specific vendor in their configuration, teams can let Twilio select the best speech provider per session.

The API makes this decision based on dynamic criteria. It evaluates the Word Error Rate for the requested language and checks current provider availability. Word Error Rate can vary significantly between providers depending on the specific language or dialect being spoken. By dynamically routing based on this metric, applications can achieve higher transcription accuracy without manual intervention.

This dynamic selection becomes the default behavior for new Real-Time Transcriptions customers. Existing customers will not see their configurations change automatically. Their static provider selections remain in place to ensure predictable behavior.

There is one notable exception to this dynamic routing and failover system. Sessions configured with a multiple language setting operate differently. If a request includes the language=multi parameter, the session will not participate in Auto mode routing or standard failover. These specific sessions continue to run exclusively on Deepgram using the Nova-3 multilingual model.

What Teams Should Check and Change

Engineering teams should review their voice application configurations to understand how these changes affect their production systems. Because failover was enabled automatically on October 22 for existing accounts, your application might already be receiving payloads from an alternate provider during primary provider downtime.

First, verify if your application requires strict feature parity. If your system logic breaks without profanity filtering or exact language variants, you might prefer an explicit error over a downgraded transcript. In that scenario, you can opt out of the failover behavior per sub-account in the Twilio Console.

Second, consider opting into Auto mode to improve baseline transcription quality. Existing customers must explicitly update their code to use this feature. You can enable it by setting the transcription engine parameter to auto on the TwiML noun or within the Calls Transcriptions API requests.

Text
<Response>
  <Start>
    <Transcription transcriptionEngine="auto" />
  </Start>
  <Dial>+1234567890</Dial>
</Response>

Third, audit your payload parsing logic. Ensure your code checks for the existence of optional fields before attempting to use them. A missing word timing array should trigger a safe fallback in your user interface, not a server error. Defensive programming around third-party payloads is essential when those payloads can change based on automated fallback routing.

Relying on third-party APIs means accepting that underlying services will occasionally fail. Moving the failover logic from the application layer to the API platform reduces the amount of custom code teams must maintain. It allows developers to focus on building core product features rather than managing vendor redundancy and payload normalization. Knowing exactly when these platform behaviors change is critical for maintaining stable software. Breakwatch maps which files in your repos call each vendor API, watches those vendors' changelogs, docs and API specs, and alerts you when a change affects your code.

Next step

Know which of your files call each vendor API

Breakwatch scans your GitHub repos, maps every vendor SDK and endpoint they use, and alerts you in Slack or email when one of those APIs ships a breaking change, down to the exact file.

Follow every vendor API change on the free Breakwatch radar

Get API changes by email

Once a week: the breaking changes and upcoming deadlines, each linked to the vendor's own page. Free, no account. One-click unsubscribe.

Sources

Every fact above links to one of these. The date beside each is the page's own.

  1. Twilio Changelog, twilio.com, October 22, 2026

More from the blog

Articles describe vendors' APIs, status pages and changelogs as their sources showed them on the dates listed. Those change; check the sources before you rely on them. Names of other companies and their products belong to them; Breakwatch is not affiliated with them.