feat(openai): capture audio transcript deltas in streaming chat completions

[arnab812]

• Update _extract_streamed_openai_response to inspect delta.audio for streaming chunks
• Extract and concatenate transcript tokens into completion["content"]
• Ensure text output is captured in Langfuse dashboard traces when using audio modalities

What does this PR do?

When streaming chat completions with audio modality enabled (modalities=["text", "audio"]), OpenAI returns spoken transcript chunks inside chunk.choices[0].delta.audio["transcript"] instead of chunk.choices[0].delta.content. Currently, this leaves the extracted completion content empty in Langfuse traces. This PR updates _extract_streamed_openai_response() to capture and concatenate transcript deltas into completion["content"].

Fixes #

Type of change

• Bug fix
• New feature
• Breaking change
• Refactor
• Documentation update
• Tooling, CI, or repo maintenance

Verification

List the main commands you ran: pytest tests/unit/test_openai.py

Checklist

• I self-reviewed the diff using code_review.md.
• I added or updated tests for behavior changes.
• I updated docs, examples, or .env.template if needed.
• I did not hand-edit generated files; if generated files changed, I used the upstream regeneration path.
• I did not commit secrets or credentials.

Greptile Summary

Adds support for collecting transcript fragments from streaming OpenAI audio deltas so Langfuse observations retain the spoken response text.

• Reads delta.audio.transcript from dictionary and object-backed audio deltas.
• Concatenates transcript fragments into the existing completion content accumulator.
• Does not add automated coverage for the new streaming response shape.

Confidence Score: 4/5

The PR appears safe to merge, with the non-blocking concern that the new streamed audio-transcript path lacks regression coverage.

The extraction change is narrowly scoped and no concrete runtime defect was established, but dictionary/object handling and multi-chunk transcript accumulation are not exercised by existing tests.

Files Needing Attention: langfuse/openai.py

Prompt To Fix All With AI

### Issue 1
langfuse/openai.py:804-805
**Streaming transcript path untested**

The new object/dictionary normalization and multi-chunk transcript concatenation have no automated coverage, so regressions can silently restore empty or malformed Langfuse trace output.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{Reviews (1): Last reviewed commit: "feat(openai): capture audio transcript d..." | Re-trigger Greptile}

Context used:

• Knowledge Base — OpenAI instrumentation

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

All committers have signed the CLA.