docs: OpenAI-compatible self-hosted endpoints are already supported #863
Conversation
There was a problem hiding this comment.
Pull request overview
This PR demonstrates that the speech_recognition library already supports OpenAI-compatible API endpoints through the OPENAI_BASE_URL environment variable, addressing issue #850. The implementation leverages the OpenAI Python client's native support for custom base URLs.
- Adds a test case demonstrating OpenAI-compatible endpoint support via environment variables
- Introduces
pytest-httpserveras a new dev dependency for HTTP mocking
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
| tests/recognizers/whisper_api/test_openai_compatible.py | New test file that validates OpenAI-compatible API endpoints work by setting OPENAI_BASE_URL environment variable |
| setup.cfg | Adds pytest-httpserver to dev dependencies for the new test's HTTP mocking |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| actual = sut.recognize_openai(audio_data) | ||
|
|
||
| assert actual == "Transcription by OpenAI compatible API" | ||
| audio_data.get_wav_data.assert_called_once_with() |
There was a problem hiding this comment.
The assertion style assert_called_once_with() is inconsistent with the existing OpenAI test in test_openai.py line 35, which uses assert_called_once() (without the _with suffix when no arguments are being checked). For consistency, this should use assert_called_once() since you're verifying it was called exactly once, regardless of arguments.
| audio_data.get_wav_data.assert_called_once_with() | |
| audio_data.get_wav_data.assert_called_once() |
| pytest-randomly | ||
| respx | ||
| numpy | ||
| pytest-httpserver |
There was a problem hiding this comment.
Adding pytest-httpserver as a dev dependency creates duplication with respx, which is already used for mocking HTTP requests in the existing OpenAI tests. This increases the dependency footprint unnecessarily. Consider removing this dependency and using respx instead in the test implementation to maintain consistency.
| pytest-httpserver |
| def test_transcribe_with_openai_compatible_api(httpserver, monkeypatch): | ||
| # https://github.com/Uberi/speech_recognition/issues/850 | ||
| httpserver.expect_request( | ||
| "/v1/audio/transcriptions", | ||
| method="POST", | ||
| ).respond_with_json({"text": "Transcription by OpenAI compatible API"}) | ||
|
|
||
| monkeypatch.setenv("OPENAI_API_KEY", "EMPTY") | ||
| monkeypatch.setenv("OPENAI_BASE_URL", httpserver.url_for("/v1")) | ||
|
|
||
| audio_data = MagicMock(spec=AudioData) | ||
| audio_data.get_wav_data.return_value = b"audio_data" | ||
|
|
||
| sut = Recognizer() | ||
| actual = sut.recognize_openai(audio_data) | ||
|
|
||
| assert actual == "Transcription by OpenAI compatible API" | ||
| audio_data.get_wav_data.assert_called_once_with() |
There was a problem hiding this comment.
This test uses pytest-httpserver for mocking HTTP requests, while the existing OpenAI tests in test_openai.py use respx for the same purpose. This creates an inconsistent testing approach and adds an unnecessary additional dependency. Consider using respx instead to align with the existing test patterns and maintain consistency across the test suite.
ftnext
changed the title
1 participant
Close #850
Add test to show supporting OpenAI compatible endpoint already