Audio transcription is typed as a generic API response

[saxon134]

AudioResource.transcription and transcription_async validate /v1/audio/transcriptions as APIResponse, whose declared fields are only id/object/data/status. The test suite accesses .text on this result, and the repository's strict mypy configuration fails with: tests/test_client.py:165: "APIResponse" has no attribute "text". Add a transcription response model with a text field and return it from both transcription methods.

File: src/globalrouter/_resources.py

Line: 540

Severity: high

Summary: wrong audio transcription response type

[saxon134]

Coding Loop Plan

Refined Issue

Audio transcription methods currently return the generic APIResponse type even though the endpoint returns a transcription payload with a text field. Add a dedicated typed response model and use it for sync and async transcription calls.

Implementation Plan

Implementation Plan

Keep the change scoped to audio transcription typing. Add a dedicated model for transcription responses, wire only the transcription endpoints to that model, export it with the other public models, and add focused tests that prove both runtime parsing and static typing expose .text as a string.

Development Tasks

1. Add typed audio transcription response model
   Define a new Pydantic model in src/globalrouter/_models.py, likely named AudioTranscription, inheriting from GlobalRouterModel. Include text: str because the existing mock endpoint returns {"text": "hello"} and the bug is specifically that .text is not typed. Keep extra="allow" behavior through the base class so provider-specific fields remain accepted. Export the new model from src/globalrouter/__init__.py and add it to __all__.
   Files: src/globalrouter/_models.py, src/globalrouter/init.py
   Tests: python -m mypy src tests, python -m ruff check src tests

2. Use the transcription model for audio transcription methods
   Update src/globalrouter/_resources.py to import AudioTranscription. Change AudioResource.transcription and AudioResource.transcription_async return annotations from APIResponse to AudioTranscription, and pass AudioTranscription to request_model / request_model_async for /v1/audio/transcriptions. Do not change speech or unrelated generic endpoints.
   Files: src/globalrouter/_resources.py
   Tests: python -m mypy src tests, python -m pytest tests/test_client.py

3. Add focused runtime and typing coverage
   Extend tests/test_client.py so the existing transcription assertion stores the result, verifies it is an AudioTranscription, and verifies result.text == "hello". Add a typing-sensitive assertion or annotation such as text: str = result.text so python -m mypy src tests fails if transcription regresses to APIResponse. If async transcription has no coverage, add a small mock-transport async test for client.audio.transcription_async(...) with the same response shape.
   Files: tests/test_client.py
   Tests: python -m pytest tests/test_client.py, python -m mypy src tests, python -m ruff check src tests