Zipformer: Optimized for Qualcomm Devices
Zipformer streaming ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text streamingly. This model is based on the transformer architecture and has been optimized for edge inference by replacing linear layers with convolutional (conv) layers. It exhibits robust performance in realistic, noisy environments, making it highly reliable for real-world applications. Specifically, it excels in long-form transcription, capable of accurately transcribing audios. Time to the first token is the encoder's latency, while time to each additional token is joiner's latency, where we assume a max decoded length specified below.
This is based on the implementation of Zipformer 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 |
|---|---|---|---|---|
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.45 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit Zipformer 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 Zipformer on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.speech_recognition
Model Stats:
- Model checkpoint: pfluo/k2fsa-zipformer-chinese-english-mixed
- Input resolution: 80x71 (0.71 seconds audio)
- Max decoded sequence length: 200 tokens
- Number of parameters (encoder): 63.2M
- Model size (encoder) (float): 242 MB
- Number of parameters (decoder): 3.47M
- Model size (decoder) (float): 13.2 MB
- Number of parameters (joiner): 3.21M
- Model size (joiner) (float): 12.2 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.048 ms | 0 - 10 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Mobile | 0.048 ms | 0 - 11 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 0.038 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 0.152 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 0.152 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.057 ms | 0 - 6 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.075 ms | 0 - 8 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 0.154 ms | 0 - 3 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.048 ms | 0 - 11 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Gen 5 Mobile | 0.054 ms | 0 - 10 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Mobile | 0.052 ms | 0 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X2 Elite | 0.034 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 0.156 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 0.156 ms | 7 - 7 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Gen 3 Mobile | 0.061 ms | 0 - 6 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS8550 (Proxy) | 0.073 ms | 0 - 8 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS9075 | 0.151 ms | 0 - 3 MB | NPU |
| decoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite For Galaxy Mobile | 0.052 ms | 0 - 7 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.042 ms | 0 - 10 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Mobile | 0.041 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 0.133 ms | 0 - 0 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 0.147 ms | 0 - 0 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 0.147 ms | 0 - 0 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 0.05 ms | 0 - 8 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 0.301 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 0.07 ms | 0 - 1 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.206 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.206 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.206 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 0.123 ms | 0 - 2 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 0.097 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 0.301 ms | 0 - 9 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 0.336 ms | 0 - 5 MB | NPU |
| decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.041 ms | 0 - 9 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 4.821 ms | 10 - 20 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Mobile | 5.459 ms | 10 - 21 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 5.345 ms | 142 - 142 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 9.829 ms | 142 - 142 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 9.829 ms | 142 - 142 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 6.565 ms | 9 - 17 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 9.122 ms | 0 - 159 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 10.452 ms | 8 - 12 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 5.459 ms | 10 - 21 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Gen 5 Mobile | 7.148 ms | 10 - 20 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Mobile | 7.53 ms | 9 - 16 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X2 Elite | 8.433 ms | 75 - 75 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 13.447 ms | 75 - 75 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 13.447 ms | 75 - 75 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Gen 3 Mobile | 11.413 ms | 9 - 17 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS8550 (Proxy) | 15.855 ms | 9 - 10 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS9075 | 16.676 ms | 8 - 12 MB | NPU |
| encoder | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite For Galaxy Mobile | 7.53 ms | 9 - 16 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 4.807 ms | 3 - 12 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Mobile | 5.463 ms | 3 - 15 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 6.067 ms | 3 - 3 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 9.514 ms | 3 - 3 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 9.514 ms | 3 - 3 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 6.548 ms | 2 - 12 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 23.477 ms | 3 - 11 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 8.797 ms | 3 - 4 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 10.581 ms | 3 - 11 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 10.581 ms | 3 - 11 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 10.581 ms | 3 - 11 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 10.15 ms | 3 - 7 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 12.087 ms | 2 - 13 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 23.477 ms | 3 - 11 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 13.077 ms | 3 - 8 MB | NPU |
| encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 5.463 ms | 3 - 15 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.132 ms | 0 - 10 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Mobile | 0.133 ms | 0 - 7 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 0.089 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 0.252 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 0.252 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.156 ms | 0 - 7 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.192 ms | 0 - 9 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 0.271 ms | 0 - 3 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.133 ms | 0 - 7 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Gen 5 Mobile | 0.136 ms | 0 - 10 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite Mobile | 0.138 ms | 0 - 7 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X2 Elite | 0.087 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 0.268 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® X Elite | 0.268 ms | 6 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Gen 3 Mobile | 0.161 ms | 0 - 6 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS8550 (Proxy) | 0.206 ms | 0 - 8 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Qualcomm® QCS9075 | 0.277 ms | 0 - 3 MB | NPU |
| joiner | PRECOMPILED_QNN_ONNX | mixed | Snapdragon® 8 Elite For Galaxy Mobile | 0.138 ms | 0 - 7 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.132 ms | 0 - 10 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Mobile | 0.146 ms | 0 - 13 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 0.177 ms | 0 - 0 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 0.254 ms | 0 - 0 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 0.254 ms | 0 - 0 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 0.139 ms | 0 - 8 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 0.487 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 0.184 ms | 0 - 1 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.319 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.319 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 0.319 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 0.236 ms | 0 - 2 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 0.229 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 0.487 ms | 0 - 9 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 0.458 ms | 0 - 5 MB | NPU |
| joiner | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.146 ms | 0 - 13 MB | NPU |
License
- The license for the original implementation of Zipformer 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.
