From d1a00001f92237cb2b27e1b3d623c49178f38caf Mon Sep 17 00:00:00 2001 From: Pete Mo <175202887+cabbage-ice-cream@users.noreply.github.com> Date: Tue, 21 Jul 2026 16:39:34 +0100 Subject: [PATCH] Address review feedback on Smart formatting (PR #286) - Drop "with variable results" from the off-list languages note; "best-effort" already conveys the variability (per review comment) - Add a note that entity output (enable_entities, spoken_form/written_form) is Standard/Enhanced only, and link to the Melia 1 feature gap in Models The larger Melia rewrite (smart formatting on by default for all languages, removing enable_entities and spoken/written form) is deferred to DEL-34274, pending Melia becoming the default and engineering sign-off. DEL-33629, DEL-33895 Co-Authored-By: Claude Opus 4.8 --- docs/speech-to-text/formatting.mdx | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/docs/speech-to-text/formatting.mdx b/docs/speech-to-text/formatting.mdx index a585e9ce..3b885bb3 100644 --- a/docs/speech-to-text/formatting.mdx +++ b/docs/speech-to-text/formatting.mdx @@ -245,6 +245,10 @@ To include detailed entity information in your JSON output, add `enable_entities By default, `enable_entities` is `false`. When enabled, entity metadata appears only in JSON output (SRT and TXT formats remain unchanged). +:::note +Entity output (`enable_entities` and the `spoken_form` and `written_form` representations) is available with the Standard and Enhanced models. The Melia 1 multilingual model does not support entity detection yet; see [Models](/speech-to-text/models). +::: + ### Output The JSON output will include: @@ -374,7 +378,7 @@ Smart formatting has had dedicated work for consistent results in these language - Swedish - Tamil & English (bilingual) -Other languages still format numbers and entities on a best-effort basis through the model, with variable results. If you rely on formatting for a language that isn't listed, test it with representative audio rather than assuming full coverage. +Other languages still format numbers and entities on a best-effort basis through the model. If you rely on formatting for a language that isn't listed, test it with representative audio rather than assuming full coverage. Formatting coverage isn't reported by [feature discovery](/speech-to-text/features/feature-discovery), which covers transcription, translation, and language identification. This page is the reference for formatting language support.