Skip to main content
Retell selects a speech recognition provider for your agent’s languages when you use a preset transcription mode. Custom Settings lets you choose a supported provider for the whole agent or for a conversation flow node. These settings apply to cascading voice agents. Speech-to-speech models handle audio directly and do not use these transcription controls.

​
Provider overview

Use the language support table to check individual languages and multilingual agent guidance to choose a combination. The dashboard filters custom providers to those that support every selected language. Transcription latency depends on the provider, language, endpointing, and the caller’s speech. Compare providers with representative web test calls; a single latency figure does not describe every configuration.

​
How Retell picks a provider

Transcription Mode determines how the provider is selected:
  • Optimize for speed uses Retell’s provider configuration for responsive transcription in the selected languages.
  • Optimize for accuracy uses the accuracy-oriented configuration for those languages.
  • Custom Settings tries your chosen provider first, using its configured endpointing where supported.
Presets can route calls across multiple providers. (Provider Config) beside a preset shows the current provider information, including routing shares when applicable. Retell can fall back to another provider if the selected provider fails. Choosing a custom provider does not disable fallback.

​
Choose a provider for one node

A conversation flow can use a global transcription mode and a different mode in an individual conversation node or subagent node. For example, use a custom provider in a node that collects account numbers, then return to the global preset for the rest of the call. Open the node’s Node settings, enable Transcription Mode, and select Custom Settings to choose its provider. Follow node transcription overrides for setup, inheritance behavior, and endpointing controls.