Skip to content

Deepgram's summary and topics silently do nothing for non-English projects #598

Description

@maboa

We always request summarize=v2&topics=true (js/hyperaudio-lite-editor-deepgram.js:329), but both features are English only. Deepgram returns an unsupported_language warning — "Topics is only supported for English." — and our extractors simply yield an empty string and an empty array. The user sees an empty summary and no topics with no indication why, and texts.summary / texts.topics are saved empty into the container.

Summarization has a second silent condition: it needs more than 50 words. Below that it returns the original text rather than a summary, and isn't billed.

Nothing here is broken as such — the transcript is unaffected — but "empty and unexplained" is the wrong answer for a feature the format advertises. Options, roughly in order of effort: don't request them when the language isn't English; surface the returned warning where the empty fields appear; or use AssemblyAI's iab_categories for topics in those languages, which covers nine (see the companion issue).

Worth knowing for later: Deepgram supports up to 100 custom topics via custom_topic, in strict or extended mode — a real hook if project- or host-supplied vocabularies ever become interesting.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions