Audio transcription response is typed as missing text

[saxon134]

AudioResource.transcription validates /v1/audio/transcriptions into APIResponse, whose declared fields do not include the endpoint's top-level text field. The repository's own test accesses .text, and mypy src tests fails with "APIResponse" has no attribute "text". Add a transcription response model with text: str and return it from both sync and async transcription methods.

File: src/globalrouter/_resources.py

Line: 540

Severity: medium

Summary: wrong return type for audio transcription

[saxon134]

Coding Loop Plan

Refined Issue

Audio transcription methods currently return APIResponse in the SDK typing, but transcription responses include a top-level text field. Type checkers therefore report client.audio.transcription(...).text as missing.

Implementation Plan

Implementation Plan

Add a dedicated typed response model for audio transcription responses, wire the sync and async audio transcription methods to return it, and add runtime plus static typing regression coverage. Keep the change scoped to the audio transcription response type; do not alter speech or unrelated generic APIResponse behavior.

Development Tasks

1. Add audio transcription response model
   Create a new Pydantic model in src/globalrouter/_models.py for audio transcription responses. It should inherit from GlobalRouterModel and declare the top-level text field returned by /v1/audio/transcriptions. Use str for text unless existing API docs or fixtures show the field can be absent. Export the new model from src/globalrouter/init.py so users can import it consistently with the other public response models.
   Files: src/globalrouter/_models.py, src/globalrouter/init.py
   Tests: python -m pytest tests/test_client.py -q

2. Return the new model from audio transcription methods
   Update src/globalrouter/_resources.py to import the new audio transcription model. Change AudioResource.transcription and AudioResource.transcription_async return annotations and request_model/request_model_async model arguments from APIResponse to the new model. Leave AudioResource.speech unchanged.
   Files: src/globalrouter/_resources.py
   Tests: python -m pytest tests/test_client.py -q, python -m mypy src

3. Add regression coverage for typed transcription text
   Extend tests to cover the new typing behavior. Keep the existing mocked /v1/audio/transcriptions response with {"text": "hello"}. Add or update assertions that the sync transcription result exposes text at runtime. Add an async transcription assertion if async audio coverage exists nearby or can be added with the existing MockTransport pattern. Add a small mypy regression file under tests/typing or an equivalent existing typing-test location that assigns client.audio.transcription(...).text to a str variable, then include a concrete command to type-check it with MYPYPATH=src.
   Files: tests/test_client.py, tests/typing/audio_transcription_response.py
   Tests: python -m pytest tests/test_client.py -q, MYPYPATH=src python -m mypy tests/typing/audio_transcription_response.py