fix: handle SSML phoneme attribute order in ssml_to_deepgram - #741
Merged
GregHolmes merged 3 commits intoSep 1, 2026
Merged
Conversation
ssml_to_deepgram only matched <phoneme> tags with alphabet before ph, so <phoneme ph="..." alphabet="ipa">word</phoneme> silently lost its pronunciation and fell through to the generic tag strip, leaving the bare word. SSML attribute order is not significant, so match the attributes as a group and pull ph out of it. Add a regression test.
NishchayMahor
requested review from
GregHolmes,
deepgram-kiley and
dg-coreylweathers
as code owners
July 5, 2026 06:58
GregHolmes
approved these changes
Sep 1, 2026
GregHolmes
added a commit
that referenced
this pull request
Sep 3, 2026
🤖 I have created a release *beep* *boop* --- ## [7.8.1](v7.8.0...v7.8.1) (2026-09-03) ### Bug Fixes * **TextBuilder:** `ssml_to_deepgram()` now preserves a `<phoneme>` pronunciation when its valid `ph` and `alphabet` attributes appear in either order. ([#741](#741)) ([7fd4b63](7fd4b63)) * **Credentials:** Explicitly passing `api_key=None` continues to disable ambient `DEEPGRAM_API_KEY` lookup, which is important for multi-tenant and test environments. ([#778](#778)) ([e675990](e675990)) * **Credentials:** `DeepgramClient()` and `AsyncDeepgramClient()` now resolve `DEEPGRAM_API_KEY` when constructed, so `load_dotenv()` can run after importing the SDK. Closes [#734](#734). ([#767](#767)) ([ec362ec](ec362ec)) * **Custom transports:** Speak V2 WebSocket connections now honor `transport_factory`, matching the routing behavior of other WebSocket APIs for proxies, test doubles, and custom-hosted transports. ([#766](#766)) ([0980663](0980663)) ### Documentation * **Transcription:** Clarified that Nova-3 assumes English when `language` is omitted; non-English and multilingual audio require an explicit language such as `fr` or `multi`. ([#771](#771)) ([4574337](4574337)) * **Examples:** Added Listen V1 live microphone transcription with optional `sounddevice`, device selection, bounded audio buffering, transcript output, and clean Ctrl-C shutdown. ([#780](#780)) ([08f0471](08f0471)) * **Examples:** Added a resilient Listen V1 live transcription pattern with exponential backoff, reconnect-aware audio buffering, timestamp continuity, and clean shutdown. ([#776](#776)) ([96b2d11](96b2d11)) * **Examples:** Added an application-owned Voice Agent session recorder that serializes received transcripts, function calls, and latency reports as JSON while leaving consent, redaction, retention, and storage policy to the application. Closes [#775](#775). ([#781](#781)) ([30ad152](30ad152)) * **Text-to-Speech:** Corrected streaming synthesis snippets to iterate the response byte chunks instead of accessing a nonexistent `.stream` attribute. ([#749](#749)) ([178724e](178724e)) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Greg Holmes <greg.holmes@deepgram.com>
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
2 participants
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.
What
ssml_to_deepgram()only recognized<phoneme>tags when the attributes were inalphabet-before-phorder. Since SSML attribute order is not significant, a perfectly valid tag written the other way was silently dropped:The regex hardcoded
alphabet=... ph=...; when reversed, the tag didn't matchphoneme_patternand fell through to the generic<[^>]+>strip, leaving only the bare word — no error, so the caller never knows the pronunciation was dropped.Fix
Match the
<phoneme>attributes as a group and pullphout of it, so order no longer matters (src/deepgram/helpers/text_builder.py). If there's nophattribute, it degrades gracefully to the plain word. Behavior for the existingalphabet-first form is unchanged.Testing
Added
test_phoneme_attribute_order_independenttoTestSsmlToDeepgram.pytest tests/custom/test_text_builder.py→ 45 passed. Pure string transformation — no API key/network.