Repository navigation
Clarify that language hints restrict labels, not just bias - #349
Open
giorgosHadji wants to merge 1 commit into
Open
giorgosHadji wants to merge 1 commit into
giorgosHadji wants to merge 1 commit into
Conversation
The Batch and Realtime input pages, and the Realtime API spec, said language hints bias detection "without restricting it" and that the model can still label languages you did not hint. That is wrong for Melia 1 and Oak 1: language labels are structurally restricted to the hinted set, so hinting ["nl", "fr"] means German or English speech is labelled nl or fr (DEL-37132, rejected as by-design). Hints also bias transcription toward the hinted languages, which reduces but does not eliminate output in other languages. The pages now describe both effects separately, based on Daniel Cochrane's suggested wording on the ticket, and add a note to omit hints when the audio's languages can't be predicted. The Batch section also now mentions Oak 1 alongside Melia 1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
mnemitz
approved these changes
Nov 12, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Batch and Realtime input pages, and the Realtime API spec, said language hints bias detection "without restricting it" and that the model can still label languages you did not hint. That is wrong for Melia 1 and Oak 1: language labels are structurally restricted to the hinted set, so hinting ["nl", "fr"] means German or English speech is labelled nl or fr (rejected as by-design).
Hints also bias transcription toward the hinted languages, which reduces but does not eliminate output in other languages. The pages now describe both effects separately.