Task4: Add OpenAI Whisper API transcriber support#34
Conversation
Implements OpenAITranscriber as an alternative transcription backend alongside existing FasterWhisper and Groq options. Changes: - Added OpenAITranscriber class in src/transcriber.py following the same pattern as GroqTranscriber - Includes API key validation, preflight checks, and retry logic - Updated TranscriberFactory to return OpenAITranscriber for 'openai' backend - Added comprehensive test coverage for OpenAITranscriber - Tests include API call mocking, segment parsing, and error handling The implementation uses OpenAI's whisper-1 model with verbose JSON response format to get segment and word-level timestamps, matching the capabilities of other transcriber backends.
Updated documentation to reflect the new OpenAI transcriber backend: - .env.example: Clarified WHISPER_BACKEND options and added note about cloud APIs - README.md: Updated transcription options to list OpenAI as official cloud option instead of "fallback", and updated technology stack table - CLOUD_INFERENCE_OPTIONS.md: Added comprehensive OpenAI Whisper API section with setup instructions, pricing ($0.006/min), features, and configuration examples The documentation now properly presents all three transcription options: 1. Local (faster-whisper) - free, runs on your hardware 2. Groq API - free cloud option with generous tier 3. OpenAI API - paid cloud option with official support
Summary of ChangesHello @Gambitnl, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request significantly enhances the application's transcription capabilities by integrating the OpenAI Whisper API. This new backend provides users with a high-quality, cloud-based option for audio transcription, complementing the existing local and Groq-based solutions. The changes include the implementation of a dedicated OpenAI transcriber, updates to configuration, and thorough documentation to guide users on setup and usage, offering greater flexibility and choice in transcription services. Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request successfully adds support for the OpenAI Whisper API as a transcription backend. The changes include the core implementation in src/transcriber.py, comprehensive unit tests, and updates to documentation in README.md and CLOUD_INFERENCE_OPTIONS.md. The code is generally well-written and follows good practices like using try...finally for resource cleanup. My main feedback focuses on improving maintainability by reducing code duplication and making the preflight checks more robust. Overall, this is a great addition to the project.
| ) | ||
|
|
||
|
|
||
| class OpenAITranscriber(BaseTranscriber): |
There was a problem hiding this comment.
There is significant code duplication between this new OpenAITranscriber and the existing GroqTranscriber. Both classes handle temporary file creation, API calls with file objects, response parsing, and cleanup in a very similar way. This duplication increases maintenance overhead, as any bug fix or improvement would need to be applied in two places.
Consider introducing a new abstract base class, for example CloudAPITranscriber, that inherits from BaseTranscriber and abstracts this common logic. This new base class could implement the transcribe_chunk method to handle the file I/O and response parsing, while leaving an abstract _make_api_call method for subclasses to implement.
This refactoring would greatly reduce code duplication and make the cloud-based transcribers easier to maintain and extend in the future.
| # Test API with minimal request | ||
| response = self.client.chat.completions.create( | ||
| messages=[{"role": "user", "content": "test"}], | ||
| model="gpt-3.5-turbo", | ||
| max_tokens=1, | ||
| ) | ||
| self.logger.debug("OpenAI API preflight check passed") |
There was a problem hiding this comment.
The preflight check currently verifies the API key by making a call to the chat.completions endpoint. While this confirms the key is valid, it doesn't guarantee that it has the necessary permissions for the audio transcription API. A more robust check would be to test the audio.transcriptions endpoint directly.
You can do this by sending a very short, silent audio clip. This ensures that the key is not only authentic but also authorized for the specific service this class uses.
| # Test API with minimal request | |
| response = self.client.chat.completions.create( | |
| messages=[{"role": "user", "content": "test"}], | |
| model="gpt-3.5-turbo", | |
| max_tokens=1, | |
| ) | |
| self.logger.debug("OpenAI API preflight check passed") | |
| # Test API with minimal audio request | |
| import io | |
| import soundfile as sf | |
| import numpy as np | |
| # Create a silent 0.1s audio clip to test the transcription endpoint directly | |
| sample_rate = 16000 | |
| silent_audio = np.zeros(int(0.1 * sample_rate), dtype=np.float32) | |
| buffer = io.BytesIO() | |
| sf.write(buffer, silent_audio, sample_rate, format='WAV') | |
| buffer.seek(0) | |
| buffer.name = "preflight.wav" # The OpenAI SDK requires a name attribute on the file-like object | |
| # Test API with minimal audio request | |
| self.client.audio.transcriptions.create( | |
| file=buffer, | |
| model="whisper-1", | |
| ) | |
| self.logger.debug("OpenAI API preflight check passed") |
2 participants
No description provided.