Silero-VAD: Optimized for Qualcomm Devices
Silero VAD is a compact, production-ready Voice Activity Detection model trained on a large multilingual corpus. It processes 32 ms audio chunks at 16 kHz and outputs a speech probability per chunk, enabling real-time detection of speech segments in streaming or file-based audio. The model uses an LSTM-based architecture and is well-suited for edge deployment on Qualcomm Snapdragon devices.
This is based on the implementation of Silero-VAD found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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 2.50, ONNX Runtime 1.30.0 | Download |
| ONNX | w8a16_mixed_int16 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.50 | Download |
| QNN_DLC | w8a16_mixed_int16 | Universal | QAIRT 2.50 | Download |
| TFLITE | float | Universal | QAIRT 2.50 | Download |
For more device-specific assets and performance metrics, visit Silero-VAD on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models 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 Silero-VAD on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.voice_activity_detection
Model Stats:
- Input resolution: 512 samples (32 ms at 16 kHz)
- Model size (float): 2.27 MB
- Number of parameters: 0.24M
- Supported sample rates: 16000 Hz
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| Silero-VAD | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.069 ms | 0 - 24 MB | NPU |
| Silero-VAD | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.07 ms | 0 - 24 MB | NPU |
| Silero-VAD | ONNX | float | Snapdragon® X2 Elite | 0.063 ms | 1 - 1 MB | NPU |
| Silero-VAD | ONNX | float | Snapdragon® X Elite | 0.072 ms | 0 - 0 MB | NPU |
| Silero-VAD | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.078 ms | 0 - 35 MB | NPU |
| Silero-VAD | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 0.126 ms | 0 - 35 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 0.334 ms | 0 - 4 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.097 ms | 0 - 29 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® QCS8450 | 0.126 ms | 0 - 35 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 0.214 ms | 0 - 3 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.072 ms | 0 - 0 MB | NPU |
| Silero-VAD | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.07 ms | 0 - 24 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.071 ms | 0 - 26 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.078 ms | 0 - 25 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.057 ms | 1 - 1 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® X Elite | 0.073 ms | 0 - 0 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.084 ms | 0 - 34 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-8275 | 0.165 ms | 0 - 4 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.099 ms | 0 - 2 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-9075 | 0.195 ms | 0 - 3 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-X7181 | 0.073 ms | 0 - 0 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-6690 | 0.28 ms | 0 - 24 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-7790 | 0.088 ms | 0 - 24 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-8750 | 0.078 ms | 0 - 25 MB | NPU |
| Silero-VAD | ONNX | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 0.088 ms | 0 - 24 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.064 ms | 0 - 24 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.065 ms | 0 - 28 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® X2 Elite | 0.165 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® X Elite | 0.174 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.067 ms | 0 - 33 MB | NPU |
| Silero-VAD | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 0.116 ms | 0 - 34 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 0.239 ms | 0 - 3 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.091 ms | 0 - 1 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® SA8775P | 0.236 ms | 0 - 26 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® SA8650P | 0.236 ms | 0 - 26 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® SA8255P | 0.236 ms | 0 - 26 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® QCS8450 | 0.116 ms | 0 - 34 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 0.152 ms | 2 - 4 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.174 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.065 ms | 0 - 28 MB | NPU |
| Silero-VAD | QNN_DLC | float | Qualcomm® SA8295P | 0.237 ms | 0 - 21 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.065 ms | 0 - 23 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Elite For Galaxy Mobile | 0.064 ms | 0 - 23 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X2 Elite | 0.149 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® X Elite | 0.177 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 8 Gen 3 Mobile | 0.079 ms | 0 - 32 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-8275 | 0.168 ms | 0 - 3 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.097 ms | 0 - 1 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA8650P | 0.237 ms | 0 - 26 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA8255P | 0.237 ms | 0 - 26 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-9075 | 0.152 ms | 2 - 4 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ IQ-X7181 | 0.177 ms | 0 - 0 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-6690 | 0.19 ms | 0 - 22 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-7790 | 0.079 ms | 0 - 23 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® Dragonwing™ Q-8750 | 0.064 ms | 0 - 23 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Qualcomm® SA7255P | 0.35 ms | 0 - 24 MB | NPU |
| Silero-VAD | QNN_DLC | w8a16_mixed_int16 | Snapdragon® 7 Gen 4 Mobile | 0.079 ms | 0 - 23 MB | NPU |
| Silero-VAD | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.071 ms | 0 - 24 MB | NPU |
| Silero-VAD | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.065 ms | 0 - 23 MB | NPU |
| Silero-VAD | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.08 ms | 0 - 31 MB | NPU |
| Silero-VAD | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 0.116 ms | 0 - 38 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 0.15 ms | 0 - 4 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.081 ms | 0 - 1 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® QCS8450 | 0.116 ms | 0 - 38 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 0.156 ms | 0 - 3 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.065 ms | 0 - 23 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® SA7255P | 0.371 ms | 0 - 24 MB | NPU |
| Silero-VAD | TFLITE | float | Qualcomm® SA8295P | 0.246 ms | 0 - 21 MB | NPU |
License
- The license for the original implementation of Silero-VAD can be found here.
References
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
