LiteLLM: `speech()` / `aspeech()` not instrumented

[braintrust-bot]

Summary

Most of the original LiteLLM gaps from this issue are now covered on main. The Braintrust LiteLLM integration currently instruments:

• completion() / acompletion()
• responses() / aresponses()
• image_generation() / aimage_generation()
• embedding() / aembedding()
• moderation()
• transcription() / atranscription()

The remaining gap is LiteLLM text-to-speech:

• litellm.speech()
• litellm.aspeech()

Calls to those APIs still produce zero Braintrust tracing on main.

What is missing

No tracing spans are created when users call litellm.speech() or litellm.aspeech() through either wrap_litellm() or patch_litellm().

Current LiteLLM integration coverage on main includes image generation, transcription, and async embeddings. What is still missing in py/src/braintrust/integrations/litellm/ is:

• speech patchers in patchers.py
• speech wrappers in tracing.py
• speech coverage in test_litellm.py

Braintrust docs status

not_found — The Braintrust LiteLLM integration docs at https://www.braintrust.dev/docs do not mention text-to-speech support.

Upstream sources

• LiteLLM text-to-speech docs: https://docs.litellm.ai/docs/text_to_speech
• LiteLLM Python SDK exposes both speech() and aspeech()

Local repo files inspected

• py/src/braintrust/integrations/litellm/__init__.py
• py/src/braintrust/integrations/litellm/patchers.py
• py/src/braintrust/integrations/litellm/tracing.py
• py/src/braintrust/integrations/litellm/test_litellm.py
• py/noxfile.py

Relationship to existing issues

• Image generation support landed via feat(litellm): trace image generation calls #261.
• Transcription support landed via feat(litellm): trace transcription calls #264.
• Async embedding patching landed via fix(litellm): patch async embeddings for auto instrumentation #186.
• LiteLLM: rerank() / arerank() not instrumented #266 tracks LiteLLM rerank() / arerank() separately.

[AbhiPrasad]

Splitting LiteLLM rerank tracking into #266 so it can be handled separately from the other LiteLLM API-function gaps covered here.