๐ป Technology Decision - fuhui14/SWEN90017-2024-TAP GitHub Wiki
๐ป Technology decisions Records
-
language:
- Frontend: Javascript
- Backend : python 3.12.3
-
Framework:
- Frontend: React
- Backend : Django
-
Library:
- Transcription: OpenAI Whisper
- Speaker Identifier:
๐๏ธ Frontend Framework
| Framework | Introduction | Version | Pros | Cons | Will Use |
|---|---|---|---|---|---|
| React | A JavaScript library for building user interfaces. React is component-based, making it easy to create interactive UIs. | v18.3.1 | - Component-Based Architecture: Perfect for building modular and reusable components, which is ideal for integrating dynamic transcription results and user interfaces. - Seamless Django Integration: React can be easily paired with Django to create a robust full-stack application, allowing Django to handle the backend logic and React to manage the frontend UI. - Optimized for Real-Time Updates: Reactโs virtual DOM efficiently handles frequent UI updates, which is crucial when displaying real-time transcription and speaker identification results from OpenAI Whisper. - Strong Ecosystem: A vast ecosystem of libraries and tools makes it easier to extend functionalities, such as state management and routing, enhancing the overall application experience. | - Requires learning JSX, which can be confusing for new users. - Often requires additional libraries for state management and routing. | x |
| Angular | A platform for building mobile and desktop web applications. Angular provides a full framework with built-in features. | v18.0.0 | - Comprehensive: Offers a full-fledged MVC framework. - Strong type-checking with TypeScript. - Two-way data binding simplifies UI updates. | - Steeper learning curve. - Can be heavy for simple applications. | |
| Vue.js | A progressive JavaScript framework for building user interfaces, focusing on simplicity and flexibility. | v3.4.38 | - Simplicity: Easy to integrate and get started. - Flexible and incrementally adoptable. - Strong community and ecosystem. | - Smaller community than React and Angular. - Somewhat less opinionated, leading to different coding practices. | |
| Svelte | A radical new approach to building user interfaces, compiling components to highly efficient imperative code. | v4.2.19 | - Performance: Compiles to highly optimized vanilla JS, eliminating the need for a virtual DOM. - Simple and clean syntax. - Reduces boilerplate code. | - Smaller community and ecosystem compared to React, Angular, or Vue. - Limited tooling and ecosystem support. | |
| Ember.js | A framework for building ambitious web applications. Ember follows a convention-over-configuration philosophy. | v5.11.0 | - Convention-Over-Configuration: Reduces decision fatigue by providing strong conventions. - Comprehensive tooling and strong community support. | - Less flexible and heavier than other frameworks. - Slower learning curve due to its comprehensive nature. |
๐ Backend Framework
| Framework | Introduction | Version | Pros | Cons | Will Use |
|---|---|---|---|---|---|
| Django | A high-level Python web framework that encourages rapid development and clean, pragmatic design. | - Integration-Friendly: Django's robust ecosystem allows seamless integration with AI libraries like OpenAI Whisper. - Batteries-included: Comes with built-in features for admin interface, authentication, ORM, etc. - Strong community support and extensive documentation. | - Monolithic: Can be heavyweight for small projects. - Steeper learning curve due to its comprehensive nature. | x | |
| Flask | A lightweight WSGI web application framework in Python. | - Flexibility: Easy to integrate with OpenAI Whisper and other AI libraries. - Minimalistic: Provides simplicity and ease of use. - Modular and easy to extend. | - Requires additional libraries for full-stack capabilities. - Less opinionated, which can lead to inconsistencies in large teams. | ||
| Tornado | A Python web framework and asynchronous networking library, originally developed at FriendFeed. | - Performance: Non-blocking and asynchronous, making it suitable for handling real-time audio processing with OpenAI Whisper. - Good for applications needing high concurrency. | - Steeper learning curve for asynchronous programming. - Smaller ecosystem compared to Django or Flask. | ||
| Pyramid | A flexible Python web framework suitable for both small and large applications. | - Scalability: Can scale with the project size, making it adaptable to evolving needs. - Allows easy integration with different components, including AI libraries like Whisper. | - Less out-of-the-box features compared to Django. - Requires more configuration, which can slow down initial development. | ||
| CherryPy | A minimalist Python web framework, object-oriented and designed for rapid development. | - Lightweight: Minimalist design allows direct integration with Whisper without unnecessary overhead. - Fast and easy to deploy. | - Lacks built-in features, requiring manual integration of many components. - Smaller community and less documentation. |
Different Library
| Library | Introduction | Version | Pros | Cons | Will Use |
|---|---|---|---|---|---|
| OpenAI Whisper | OpenAI Whisper is a large-scale speech recognition model designed specifically for multi-language and multi-task speech recognition and translation tasks. It uses 680,000 hours of diverse audio data for training, can accurately transcribe and translate speech in a variety of contexts and languages, and can perform well on new datasets without fine-tuning. | v20231117 | - Multi-language support: supports multiple languages โโand tasks, and is highly adaptable. - High robustness: has good generalization ability in noisy environments and datasets with different distributions. - No fine-tuning required: can achieve excellent performance in zero-shot settings. | - Data dependence: Relying on a large amount of weakly supervised data, which may contain noise and annotation errors. - Resource requirements: A large amount of computing resources are required during training and inference, especially when processing large-scale data. | x |
| pyannote.audio | pyannote.audio is an open-source toolkit written in Python for speaker diarization. Based on PyTorch machine learning framework, it comes with state-of-the-art pretrained models and pipelines, that can be further finetuned to your own data for even better performance. | version3.3.1 | - Provides pre-trained models, easy to get started. - Supports a variety of speech tasks. | - Requires higher computing resources, especially when processing long audio. - Not easy to use. | |
| speechbrain | SpeechBrain is an open-source PyTorch toolkit that accelerates Conversational AI development, i.e., the technology behind speech assistants, chatbots, and large language models. It is crafted for fast and easy creation of advanced technologies for Speech and Text Processing. | v1.0 | - Supports multi-tasking and modular design. - Has a large number of pre-trained models and rich community support. | - Requires some background knowledge in machine learning. - Configuration and optimization may be complicated. | |
| Resemblyzer | Resemblyzer allows you to derive a high-level representation of a voice through a deep learning model (referred to as the voice encoder). Given an audio file of speech, it creates a summary vector of 256 values that summarizes the characteristics of the voice spoken. | git | - Can quickly generate speaker embeddings. - Easy to integrate into other speech processing pipelines. | - Limited processing efficiency for long audio. |
๐งช Testing Tools
| Component | Testing Framework | Purpose |
|---|---|---|
| Frontend | Jest | Test JavaScript logic & React components |
| Backend | Django unittest |
Test Django models and core functions |