---
library_name: pytorch
license: other
tags:
- real_time
- android
pipeline_tag: audio-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/web-assets/model_demo.png)

# YamNet: Optimized for Qualcomm Devices

An audio event classifier trained on the AudioSet dataset to predict audio events from the AudioSet ontology employing the Mobilenet_v1 depthwise-separable convolution architecture.

This is based on the implementation of YamNet found [here](https://github.com/w-hc/torch_audioset).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yamnet) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-onnx-w8a16.zip)
| ONNX | w8a8 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yamnet/releases/v0.46.0/yamnet-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[YamNet on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/yamnet)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yamnet) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [YamNet on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yamnet) for usage instructions.

## Model Details

**Model Type:** Model_use_case.audio_classification

**Model Stats:**
- Model checkpoint: yamnet.pth
- Input resolution: 1x1x96x64
- Number of parameters: 3.73M
- Model size (float): 14.2 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| YamNet | ONNX | float | Snapdragon® X Elite | 0.286 ms | 8 - 8 MB | NPU
| YamNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.33 ms | 0 - 109 MB | NPU
| YamNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.367 ms | 0 - 117 MB | NPU
| YamNet | ONNX | float | Qualcomm® QCS9075 | 0.487 ms | 0 - 3 MB | NPU
| YamNet | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.269 ms | 0 - 92 MB | NPU
| YamNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.26 ms | 0 - 92 MB | NPU
| YamNet | ONNX | w8a16 | Snapdragon® X Elite | 0.21 ms | 4 - 4 MB | NPU
| YamNet | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.265 ms | 0 - 106 MB | NPU
| YamNet | ONNX | w8a16 | Qualcomm® QCS6490 | 11.068 ms | 9 - 14 MB | CPU
| YamNet | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.317 ms | 0 - 11 MB | NPU
| YamNet | ONNX | w8a16 | Qualcomm® QCS9075 | 0.394 ms | 0 - 3 MB | NPU
| YamNet | ONNX | w8a16 | Qualcomm® QCM6690 | 6.783 ms | 6 - 13 MB | CPU
| YamNet | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.208 ms | 0 - 96 MB | NPU
| YamNet | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 4.83 ms | 6 - 13 MB | CPU
| YamNet | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.204 ms | 0 - 95 MB | NPU
| YamNet | ONNX | w8a8 | Snapdragon® X Elite | 0.22 ms | 4 - 4 MB | NPU
| YamNet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.255 ms | 0 - 105 MB | NPU
| YamNet | ONNX | w8a8 | Qualcomm® QCS6490 | 1.655 ms | 0 - 6 MB | CPU
| YamNet | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.333 ms | 0 - 6 MB | NPU
| YamNet | ONNX | w8a8 | Qualcomm® QCS9075 | 0.393 ms | 0 - 3 MB | NPU
| YamNet | ONNX | w8a8 | Qualcomm® QCM6690 | 0.825 ms | 0 - 8 MB | CPU
| YamNet | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.207 ms | 0 - 92 MB | NPU
| YamNet | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.561 ms | 0 - 7 MB | CPU
| YamNet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.2 ms | 0 - 95 MB | NPU
| YamNet | QNN_DLC | float | Snapdragon® X Elite | 0.286 ms | 0 - 0 MB | NPU
| YamNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.181 ms | 0 - 38 MB | NPU
| YamNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.211 ms | 0 - 25 MB | NPU
| YamNet | QNN_DLC | float | Qualcomm® SA8775P | 0.36 ms | 0 - 23 MB | NPU
| YamNet | QNN_DLC | float | Qualcomm® QCS9075 | 0.259 ms | 0 - 2 MB | NPU
| YamNet | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 0.355 ms | 0 - 39 MB | NPU
| YamNet | QNN_DLC | float | Qualcomm® SA8295P | 0.556 ms | 0 - 20 MB | NPU
| YamNet | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.161 ms | 0 - 25 MB | NPU
| YamNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.15 ms | 0 - 24 MB | NPU
| YamNet | QNN_DLC | w8a16 | Snapdragon® X Elite | 0.214 ms | 0 - 0 MB | NPU
| YamNet | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.116 ms | 0 - 35 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 0.631 ms | 0 - 2 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 0.421 ms | 0 - 23 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 0.136 ms | 0 - 1 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® SA8775P | 0.268 ms | 0 - 25 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 0.184 ms | 0 - 2 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 0.502 ms | 0 - 22 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 0.214 ms | 0 - 36 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® SA7255P | 0.421 ms | 0 - 23 MB | NPU
| YamNet | QNN_DLC | w8a16 | Qualcomm® SA8295P | 0.428 ms | 0 - 21 MB | NPU
| YamNet | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.094 ms | 0 - 21 MB | NPU
| YamNet | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 0.171 ms | 0 - 22 MB | NPU
| YamNet | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.098 ms | 0 - 25 MB | NPU
| YamNet | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.202 ms | 0 - 0 MB | NPU
| YamNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.11 ms | 0 - 35 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 0.624 ms | 0 - 2 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 0.392 ms | 0 - 24 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.131 ms | 0 - 1 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.273 ms | 0 - 24 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.183 ms | 0 - 2 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 0.434 ms | 0 - 22 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.205 ms | 0 - 36 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® SA7255P | 0.392 ms | 0 - 24 MB | NPU
| YamNet | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.443 ms | 0 - 21 MB | NPU
| YamNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.094 ms | 0 - 25 MB | NPU
| YamNet | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.164 ms | 0 - 22 MB | NPU
| YamNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.089 ms | 0 - 25 MB | NPU
| YamNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.181 ms | 0 - 44 MB | NPU
| YamNet | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 1.962 ms | 0 - 20 MB | GPU
| YamNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.221 ms | 0 - 1 MB | NPU
| YamNet | TFLITE | float | Qualcomm® SA8775P | 0.368 ms | 0 - 31 MB | NPU
| YamNet | TFLITE | float | Qualcomm® QCS9075 | 0.261 ms | 0 - 10 MB | NPU
| YamNet | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 0.355 ms | 0 - 45 MB | NPU
| YamNet | TFLITE | float | Qualcomm® SA7255P | 1.962 ms | 0 - 20 MB | GPU
| YamNet | TFLITE | float | Qualcomm® SA8295P | 0.561 ms | 0 - 27 MB | NPU
| YamNet | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.159 ms | 0 - 32 MB | NPU
| YamNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.146 ms | 0 - 30 MB | NPU
| YamNet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.109 ms | 0 - 35 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCS6490 | 0.51 ms | 0 - 6 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 0.418 ms | 0 - 23 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.131 ms | 0 - 7 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® SA8775P | 0.784 ms | 0 - 24 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.182 ms | 0 - 6 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCM6690 | 0.426 ms | 0 - 21 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.208 ms | 0 - 37 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® SA7255P | 0.418 ms | 0 - 23 MB | NPU
| YamNet | TFLITE | w8a8 | Qualcomm® SA8295P | 0.434 ms | 0 - 20 MB | NPU
| YamNet | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.101 ms | 0 - 26 MB | NPU
| YamNet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.169 ms | 0 - 21 MB | NPU
| YamNet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.095 ms | 0 - 24 MB | NPU

## License
* The license for the original implementation of YamNet can be found
  [here](https://github.com/w-hc/torch_audioset/blob/master/LICENSE).

## References
* [MobileNets Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861)
* [Source Model Implementation](https://github.com/w-hc/torch_audioset)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).