Hayasuki
/

ChuckleNet-Genki

@@ -19,53 +19,151 @@ tags:
 - genki
 metrics:
 - f1
-- iou_f1
 datasets:
 - StandUp4AI
 papers:
 - https://github.com/Das-rebel/ChuckleNet
 ---
 # 🎭 ChuckleNet-Genki (元気)
-**The original spirit.** First generation ChuckleNet model for laughter detection in stand-up comedy audio.
-> *"Genki" means spirit, energy, vitality — the life force of comedy.*
 ---
-## Overview
-Genki was the first model trained on the full 620-video StandUp4AI dataset using weak labels (VTT `[laughter]` transcript markers). It established the baseline for laughter detection using frozen WavLM embeddings.
-While superseded by **ChuckleNet-Ten** (human-verified labels), Genki remains useful for:
-- Historical reference and benchmarking progress
-- Weak-label baseline comparisons
-- Fast inference on large datasets
 ---
-## Architecture
-- **Backbone:** Frozen `microsoft/wavlm-base` (768-dim)
-- **Head:** Linear classifier → 2 classes (laugh / not-laugh)
-- **Input:** 25ms audio frames → WavLM embeddings
-- **Output:** Binary laughter probability
 ---
-## Performance
-| Metric | Value |
-|--------|-------|
-| Validation F1 | 0.27 |
-| Event IoU-F1@0.2 | 0.23 |
-| Average Precision | 0.14 |
-| Downloads | **265** |
 ---
-## Quick Start
 ```python
 from transformers import pipeline
@@ -74,34 +172,50 @@ pipe = pipeline(
     "audio-classification",
     model="Hayasuki/ChuckleNet-Genki"
 )
-results = pipe("standup_clip.mp3")
 ```
 ---
-## Limitations
-⚠️ **Weak labels only** — trained on VTT transcript markers, not human-verified laughter events. Do not use for clinical or production deployment without human validation.
 ---
-## Citation
 ```bibtex
 @misc{chuckleNetGenki2026,
   author = {Subhajit Das},
   title = {ChuckleNet-Genki: First Generation Laughter Detection},
   year = {2026},
-  url = {https://huggingface.co/Hayasuki/ChuckleNet-Genki}
 }
 ```
 ---
-## Related Models
-| Model | Labels | F1 | Purpose |
-|-------|--------|-----|---------|
-| **ChuckleNet-Ten** | Human-verified | 0.975 | Primary model |
-| **ChuckleNet-Kage** | Weak (v32) | 0.33 | Research comparison |
-| **ChuckleNet-Genki** | Weak (v1) | 0.27 | Historical reference |

 - genki
 metrics:
 - f1
 datasets:
 - StandUp4AI
 papers:
 - https://github.com/Das-rebel/ChuckleNet
 ---
+<div align="center">
 # 🎭 ChuckleNet-Genki (元気)
+***"The original spirit."***
+**First-generation laughter detection model — where the journey began.**
+[![Downloads](https://img.shields.io/badge/Downloads-265+-brightgreen?style=flat-square)](https://huggingface.co/Hayasuki/ChuckleNet-Genki)
+[![Model Size](https://img.shields.io/badge/Parameters-230K-blue?style=flat-square)](https://huggingface.co/Hayasuki/ChuckleNet-Genki)
+[![License](https://img.shields.io/badge/License-MIT-green?style=flat-square)](https://opensource.org/licenses/MIT)
+[![Dataset](https://img.shields.io/badge/Dataset-StandUp4AI%20620v-purple?style=flat-square)](https://github.com/Standup4AI/dataset)
+*The most-downloaded ChuckleNet model — establishing the baseline for all that followed.*
+</div>
+---
+## 📜 The Origin Story
+```
+┌─────────────────────────────────────────────────────────────────┐
+│                    GENKI TIMELINE                               │
+├─────────────────────────────────────────────────────────────────┤
+│                                                                 │
+│   2024-09   "Can frozen WavLM detect laughter?"                 │
+│              Question posed, first experiment designed          │
+│                         │                                      │
+│                         ▼                                      │
+│   2024-10   Genki v1 trained on 620 StandUp4AI videos           │
+│              Weak labels: VTT [laughter] transcript markers      │
+│              Result: F1 = 0.27 — proof of concept!              │
+│                         │                                      │
+│                         ▼                                      │
+│   2025-01   "Weak labels are noisy — need human verification" │
+│              → ChuckleNet-Ten project begins                    │
+│                         │                                      │
+│                         ▼                                      │
+│   2025-06   ChuckleNet-Ten: F1 = 0.975 (human-verified)        │
+│              ChuckleNet-Kage: IoU-F1 = 0.229 (v32)            │
+│                                                                 │
+│   Genki lives on as the historical reference point.             │
+│                                                                 │
+└─────────────────────────────────────────────────────────────────┘
+```
+---
+## 🧬 Evolutionary Tree
+```
+ChuckleNet Family Tree
+                    ┌──────────────────┐
+                    │  microsoft/      │
+                    │  wavlm-base     │
+                    │  (frozen)       │
+                    └────────┬─────────┘
+                             │
+            ┌────────────────┼────────────────┐
+            │                │                │
+            ▼                ▼                ▼
+    ┌───────────────┐  ┌───────────────┐  ┌───────────────┐
+    │ 🎭 Genki v1   │  │ 👻 Kage v32   │  │ 🏔️ Ten       │
+    │ Weak labels   │  │ Weak labels   │  │ Human-verified│
+    │ 620 videos    │  │ 620 videos    │  │ 87 videos     │
+    │ F1 = 0.27     │  │ IoU = 0.229   │  │ F1 = 0.975    │
+    │ First attempt │  │ Refined probe │  │ Gold standard │
+    └───────────────┘  └───────────────┘  └───────────────┘
+```
+---
+## 📊 Performance Over Time
+```
+Model Evolution:
+Genki v1 (Sep 2024)   ████████░░░░░░░░░░░░░░░░░░  F1 = 0.27
+                            │
+Kage v32 (Mar 2025)   ████████████████░░░░░░░░░░░  IoU = 0.229
+                            │                        (different metric)
+                            │
+Ten (Jun 2025)        ████████████████████████████  F1 = 0.975 ✅
+                            │
+                  Human-verified labels
+                  enabled this jump
+```
 ---
+## 🔧 Technical Specs
+| Component | Specification |
+|-----------|---------------|
+| **Backbone** | microsoft/wavlm-base (frozen, 768-dim) |
+| **Head** | Linear: 768→2 |
+| **Training** | 620 StandUp4AI videos, weak labels |
+| **Optimizer** | AdamW, lr=1e-3 |
+| **Input** | 25ms audio frames → WavLM → classifier |
+| **Output** | Binary: laugh / no-laugh |
+---
+## 📈 Baseline Metrics
+| Metric | Value | Note |
+|--------|-------|------|
+| **F1 Score** | `0.27` | Primary metric |
+| **Event IoU-F1@0.2** | `0.23` | Secondary metric |
+| **Average Precision** | `0.14` | Precision-recall area |
+| **Downloads** | **265+** | Most popular ChuckleNet model |
 ---
+## 🎯 Use Cases for Genki
+| Use Case | Recommendation |
+|----------|----------------|
+| **Historical reference** | ✅ Perfect — first model, established baseline |
+| **Benchmark comparisons** | ✅ Good — track improvement over iterations |
+| **Fast prototyping** | ⚠️ Use Ten instead (better accuracy) |
+| **Production deployment** | ❌ Use Ten instead (human-verified) |
+| **Research citations** | ⚠️ Cite Genki for weak-label baseline |
 ---
+## 🆚 Comparison Table
+| Model | Labels | F1 | IoU-F1 | Downloads | Best For |
+|-------|--------|-----|--------|----------|----------|
+| 🏔️ **Ten** | Verified | **0.975** | 0.330 | New | Production ✅ |
+| 👻 **Kage** | Weak | 0.27 | **0.229** | 49 | Research |
+| 🎭 **Genki** | Weak | 0.27 | 0.23 | **265** | Historical |
 ---
+## 🚀 Quick Start
 ```python
 from transformers import pipeline
     "audio-classification",
     model="Hayasuki/ChuckleNet-Genki"
 )
+result = pipe("standup_clip.mp3")
+# [{'label': 'no_laugh', 'score': 0.77},
+#  {'label': 'laugh', 'score': 0.23}]
 ```
 ---
+## ⚠️ Limitations
+- **Weak labels only** — VTT transcript markers, not human-verified
+- **Low F1** — 0.27 is a baseline, not production-ready
+- **Linear head** — no hidden layers limits representation capacity
+- **Not for production** — use ChuckleNet-Ten instead
+---
+## 📚 Further Reading
+| Model | Description |
+|-------|-------------|
+| [🏔️ ChuckleNet-Ten](https://huggingface.co/Hayasuki/ChuckleNet-Ten) | Primary model, human-verified, F1=0.975 |
+| [👻 ChuckleNet-Kage](https://huggingface.co/Hayasuki/ChuckleNet-Kage) | Research model, IoU-F1=0.229 |
 ---
+## 📄 Citation
 ```bibtex
 @misc{chuckleNetGenki2026,
   author = {Subhajit Das},
   title = {ChuckleNet-Genki: First Generation Laughter Detection},
   year = {2026},
+  url = {https://huggingface.co/Hayasuki/ChuckleNet-Genki},
+  note = {First attempt at laughter detection using frozen WavLM on 620 StandUp4AI videos}
 }
 ```
 ---
+<div align="center">
+[![Hugging Face](https://img.shields.io/badge/🤗-Hugging%20Face-FFD21E?style=for-the-badge&logo=huggingface)](https://huggingface.co/Hayasuki/ChuckleNet-Genki)
+*The spirit that started it all. 265+ downloads and counting.*
+</div>