[pylrn]

Problem

1. TTS(language=None) crashes: In TTS.__init__, language was typed NotGivenOr[str] and processed via _language_name(language) if is_given(language) else None. When a caller explicitly passed language=None (e.g., from an agent configuration dictionary or factory defaulting optional arguments to None), is_given(None) evaluated to True, passing None to _language_name which called None.strip() and raised AttributeError / ValueError: language cannot be empty. This was also inconsistent with TTS.update_options(language=None), which explicitly supports None to reset back to the server's default language.
2. STT(language_codes=None) and STT(language_code=None) crashes: In STT.__init__, an explicit language_codes=None or language_code=None failed the _U3_PRO_MODELS check on non-U3-Pro models (because is_given(None) is True), and for U3-Pro models _normalize_language_codes attempted to iterate over None (for code in language_codes), raising TypeError: 'NoneType' object is not iterable.
3. update_options(language_codes=None) crashes: Calling update_options(language_codes=None) on STT or SpeechStream passed None directly into _normalize_language_codes, raising TypeError: 'NoneType' object is not iterable instead of clearing steering (normalized to []), unlike "" or [].

Changes

• TTS:
  • Type language parameter as NotGivenOr[str | None] = NOT_GIVEN in TTS.__init__.
  • Guard _language_name(language) with if is_given(language) and language is not None else None so explicit None leaves language to the server default (matching update_options(language=None)).
• STT:
  • Update language_code and language_codes parameter types to accept None in STT.__init__, STT.update_options, and SpeechStream.update_options.
  • In _normalize_language_codes, return [] when language_codes is None (matching "" and []).
  • In STT.__init__, treat language_code and language_codes as NOT_GIVEN when empty (including None), keeping them exempt from the non-U3-Pro model gate at construction.
• Tests:
  • Added unit test in tests/test_plugin_assemblyai_tts.py verifying TTS(language=None) leaves _opts.language as None.
  • Added unit tests in tests/test_plugin_assemblyai_stt.py verifying STT(language_codes=None), STT(language_code=None), and update_options(language_codes=None) behave as unset/cleared steering.

Testing

• Ran uv run pytest tests/test_plugin_assemblyai_tts.py tests/test_plugin_assemblyai_stt.py (178 passed).
• Ran uv run ruff check and uv run ruff format --check (clean).
• Ran uv run mypy livekit-plugins/livekit-plugins-assemblyai/ (0 issues in 5 source files).

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[devin-ai-integration]

Devin Review found 1 potential issue.

[devin-ai-integration]

🟡 Unset language blocks valid steering

With language_code=None and language_codes=['es'], STT raises a mutual-exclusion error before clearing the unset value. The reverse combination fails too, so callers passing both optional fields cannot select a language.

Learn more

The constructor treats None as an unset language option, but its mutual-exclusion check runs before these two normalization lines. is_given(None) is true, so an explicitly unset field triggers the check when the other field has a real value. The same mismatch applies in either argument order. The check and the new normalization must use the same meaning of unset.

Example: STT(api_key="k", language_code=None, language_codes=["es"]) raises ValueError instead of storing ["es"] in _opts.language_codes.

Recommended fix: Normalize both empty or None language arguments to NOT_GIVEN before the mutual-exclusion check in STT.__init__; retain the error when both normalized arguments contain actual codes.

---

Was this helpful? React with 👍 or 👎 to provide feedback.