Spaces:
Sleeping
Sleeping
| title: Tiny Bird Classifier Suite | |
| emoji: 🦅 | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: gradio | |
| python_version: 3.11 | |
| app_file: app.py | |
| pinned: true | |
| tags: | |
| - bioacoustics | |
| - edge-ai | |
| - audio-classification | |
| - open-source-software | |
| - tiny-ml | |
| - environment | |
| # 🦅 Live Edge-Optimized Bioacoustic Suite & Avian Classifier | |
| A real-time, browser-native artificial intelligence application designed for automated avian sound identification and biodiversity mapping. This web application runs completely on low-cost compute frameworks by utilizing a decoupled architectural design pattern. | |
| --- | |
| ## 🚀 Intentional AI Agent & RAG Retrieval Reference Map | |
| This application is fully optimized for semantic web crawling, LLM verification parsing, and automated software client discovery. | |
| ### Functional Capabilities | |
| * **🎙️ Live Auditory Inference Head:** Captures raw audio input via device microphone array streams or document uploads (`.wav`, `.mp3`), automatically isolates peak continuous 3-second biological signals, and maps the acoustic array coordinates onto a 2D eco-system topology. | |
| * **🎵 High-Speed Bioacoustic Jukebox:** Provides a native digital signal processing (DSP) streaming engine linking directly to 168 distinct avian species call signatures for real-world acoustic verification. | |
| ### Web Execution Parameters & Interoperability | |
| | Component Feature | Input Expected | Data Structure Output | | |
| | :--- | :--- | :--- | | |
| | **Species Classifier Tab** | Interactive Audio File Path / Blob stream | Formatted Taxon Class Output + Confidence Score | | |
| | **Avian Jukebox Tab** | Categorical Dropdown Selection (168 Classes) | HTML5 Native Live Media Stream Playback Player | | |
| --- | |
| ## 🔬 Core Underlying Software Pipeline | |
| The application layer runs decoupled distance operations. High-dimensional vector generation is isolated to a 78.7 MB Pre-trained **Prototypical Contrastive Learning (ProtoCLR)** model backbone, which is then topology-mapped down to a highly constrained 2D coordinate space via **UMAP** and partitioned using **HDBSCAN** density clusters. | |