File size: 7,264 Bytes

---
license: other
language:
- en
tags:
- on-device
- mobile
- android
- ios
- melange
- zetic
- whisper
pipeline_tag: automatic-speech-recognition
base_model:
- openai/whisper-base
---

<!-- ============================================== -->
<!-- FILL THIS IN BEFORE PUBLISHING                -->
<!-- ============================================== -->
<!-- 

🚨 Model License Rule (One Rule Applies to All)

**Never claim or imply commercial usage rights unless the base model license explicitly allows commercial use.**

- If the base model license allows commercial use → you may state “commercial use allowed (subject to the base model license)”.
- If the base model license restricts commercial use, is AGPL/GPL, research-only, or dual-licensed → **do NOT mention commercial use**. Always defer to the base model license.

ZETIC licenses only the runtime and optimization artifacts.  
Model licensing responsibility always remains with the original model owner.

--->

---

![LinkedIn Banner](https://cdn-uploads.huggingface.co/production/uploads/68c8c30d7c55656282d829fe/_l5ZggH1LMGTw0lFGgN_l.png)

## Melange: On-device AI for Mobile — *Select. Benchmark. Deploy.*  
<div align="center">

[**🌐 Website**](https://zetic.ai) | [**📚 Docs**](https://docs.zetic.ai) | [**🎯 Melange Dashboard**](https://mlange.zetic.ai) | [**💻 GitHub Apps Repo**](https://github.com/zetic-ai/ZETIC_Melange_apps/issues) | [**💬 Discord**](https://discord.gg/q6vW4UscRY)

</div>

---

## ⚡️ What is ZETIC-ai/whisper-tiny-decoder?

ZETIC-ai/whisper-tiny-encoder is a highly optimized version of **[Whisper Tiny by OpenAI](https://huggingface.co/openai/whisper-tiny)**, built to run **on-device** (Android, iOS) with **Melange SDK**. 
<br>
<br>
For the encoder model that runs locally: [Check model card](https://huggingface.co/ZETIC-ai/whisper-tiny-encoder)

**Key Details:**
- 🎯 **Task**: Automatic Speech Recognition
- 🚀 **Engine**: ZETIC Melange Runtime
- 💡 **Best For**: Speech to Text, Speech Translation

Unlike standard GGUF or ONNX exports where you manually download `.tflite` or `.gguf` files and bundle them into your APK/IPA, **this model is fetched automatically** by the Melange SDK at runtime: no asset management, no file hosting, no bloated app sizes.

---

## 🚀 How to Use

Deploy in seconds. The Melange SDK handles model fetching, NPU delegation, and hardware acceleration automatically.

### Android (Kotlin)
```kotlin
val model = ZeticMLangeHFModel(context, "ZETIC-ai/whisper-tiny-decoder")
val output = model.run(inputData)
```


### iOS (Swift)
```swift
let model = try await ZeticMLangeHFModel("ZETIC-ai/whisper-tiny-decoder")
let outputs = try model.run(inputs: inputs)
```

💡 **Looking for Java, React Native, or Flutter?** Visit [Melange Dashboard](https://mlange.zetic.ai).

**Installation**: [View docs](https://docs.zetic.ai/hugging-face)

---

## 📊 Benchmarks & Compatibility

**Optimized for Flagships.** This model release is specifically tuned for the latest flagship smartphones.

| Device | Chipset | Inference Time (ms) | 
| :--- | :--- | :--- |
| **Samsung Galaxy S25 Ultra** | Snapdragon 8 Elite | TBD | 
| **Samsung Galaxy S25** | Snapdragon 8 Elite | TBD | 
| **iPhone 16 Pro** | A18 Pro | TBD | 
| **iPhone 16** | A18 | TBD | 

🚀 **Beyond the Flagships:** Melange supports a vast ecosystem of **100+ devices**. <br><br>

Melange SDK automatically handles optimization and fallback for the wider ecosystem, including budget & legacy smartphones. <br>
**[Explore the full ecosystem optimization on Melange Dashboard]([mlange.zetic.ai])**

---

## 🤔 Why Melange?

<div>

### **1. The Old Way (Slow & Manual)**
Separate iOS/Android Workflows | Manual Updates Require Full Re-Deployment

![Frame 3](https://cdn-uploads.huggingface.co/production/uploads/68c8c30d7c55656282d829fe/eN64F5MEEaGyCVisF-4R_.png)

### **2. Semi-Auto SDKs (Hardware-Locked)**
Hardware-Locked Model Cards | Manual File Management (Drag & Drop)

![Frame 8](https://cdn-uploads.huggingface.co/production/uploads/68c8c30d7c55656282d829fe/_vTeGpdv4K11lxwmf1VD5.png)


### **3. The Melange Way (Fast & Automatic)**
Device-Aware Auto-Routing | Hardware-Agnostic NPU API | Small App Size

![Frame 10](https://cdn-uploads.huggingface.co/production/uploads/68c8c30d7c55656282d829fe/IHH9UqRaXvawZATZ1A9PN.png)

</div>

✅ **Zero Asset Management**: We host the model & the complexity. Your app calls '[MODEL_ID]', and we deliver. <br/>
✅ **Automatic Fetching**: The SDK streams the model at initialization, no manual file uploads.<br/>
✅ **Device-Aware Auto-Routing**: The SDK auto-detects and routes the best-fit runtime variant for that device.<br/>
✅ **Unified Workflow**: Maintain a single, identical deployment path for both iOS and Android. <br/>
✅ **Instant Updates**: Push model updates instantly without waiting for App Store approvals.  <br/>
✅ **Global Scale**: Optimized paths for mid-range and low-end chips, not just flagships.<br/>
<br/>

---

## 🚀 Unlock Full Performance

**Advanced Benchmark Reports • Extended Model Catalog • 100+ Optimized Devices**

<a href="https://mlange.zetic.ai">
  <img src="https://img.shields.io/badge/🚀_Open_Dashboard-34A9A3?style=for-the-badge" alt="MeLange Dashboard" height="30" />
</a>

---

## 📄 License & Terms

**Base Model**  
Model weights are licensed under the original license provided by the model author.  
See the upstream model card for full terms.

**ZETIC Hugging Face Runtime Bundle**  
This release includes a pre-optimized runtime + model bundle enabling local, on-device inference across compatible devices.

The runtime is optimized and benchmarked for a limited set of flagship devices.  
On other compatible devices, the model may still run, but performance may vary.

Advanced device coverage and performance guarantees are available via the **Melange Dashboard (Pro)**.

**You may**
- Run the model locally on compatible devices
- Embed it in applications distributed to end users
- Use it in accordance with the base model license

**You may not**
- Extract or reuse the runtime outside this package
- Reverse engineer the optimization format
- Redistribute the runtime as a standalone SDK or service
- Bypass license verification or telemetry (if/when enabled)

**More devices, custom models, and performance guarantees**  
Get started at the **Melange Dashboard**: https://mlange.zetic.ai  

Enterprise or custom licensing: contact@zetic.ai

---
## 📞 Support & Community

- **📚 Documentation**: [https://docs.zetic.ai](https://docs.zetic.ai)
- **🐛 Melange Apps Repo**: [https://github.com/zetic-ai/ZETIC_Melange_apps/](https://github.com/zetic-ai/ZETIC_Melange_apps/)
- **💬 Discord**: [https://discord.gg/q6vW4UscRY](https://discord.gg/q6vW4UscRY)

---

<div align="center">

**By ZETIC.ai**  
*On-Device AI for Everything — for any model, on any device, in any framework*

[⭐ Star us on GitHub](https://github.com/zetic-ai/ZETIC_Melange_apps) • [👾 Discord](https://discord.gg/q6vW4UscRY) • [🚀 Try Melange](https://mlange.zetic.ai) • [📖 Documentation](https://docs.zetic.ai)

</div>](https://www.notion.so/Melange-Dashboard-2-0-Release-Checklist-2fe933a34b9880bd9494c31ea06d4f08?source=copy_link)