Skip to main content
Code switching is when speakers change language during a conversation between turns. Gladia’s multilingual transcription handles code switching with solaria-1 when language_config.code_switching is set to true: it uses language detection on each utterance, transcribes it in that language, and returns the detected language code with each utterance and word. We recommend listing the languages you expect in language_config.languages, chosen from the supported languages, for better accuracy. No spoken content is lost when speakers switch languages. To get the full transcript in another language, use translation. For background on how code switching works, see what is code switching.
For best accuracy and latency, provide a small set of expected languages. Avoid listing dozens of similar languages.

​
How to use code switching ?

Enable code switching in the session configuration:
When enabled, transcript messages include the detected language per utterance and per word. Use the transcript message schema for details.
Do not enable code_switching with an empty languages list. When no languages are specified, the language detector evaluates every utterance against 100+ supported languages, leading to frequent misdetections - especially between similar-sounding languages.

​
Recommendations

  • Limit the number of languages to those you expect.
  • Combine with custom vocabulary per language for domain terms.
  • Use supported languages to pick valid ISO 639-1 codes.
  • Check model selection for language and code-switching availability by model. Live transcription uses Solaria-1.
  • Multiple languages or code switching can reduce partial transcript accuracy, keep the expected-language list focused.
  • Supported languages describe coverage, not guaranteed accuracy on your production audio. Review async benchmarks with their model and dataset limits, then evaluate with the blind comparison tool and pricing.