Hugging Face's logo

qualcomm
/
Zipformer

Automatic Speech Recognition
PyTorch
foundation
android
Model card Files
xet
 Community

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® X2 Elite QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Snapdragon® X Elite QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 1 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ QCS8550 (Proxy) QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-9075 QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 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 Snapdragon® 8 Gen 1 Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8550 (Proxy) QAIRT 2.45 Download
QNN_CONTEXT_BINARY float Qualcomm® SA8775P QAIRT 2.45 Download
QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-9075 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 Snapdragon® 8 Elite Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 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® X2 Elite 0.033 ms 1 - 1 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® X Elite 0.044 ms 7 - 7 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile 0.062 ms 0 - 7 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 1 Mobile 0.1 ms 0 - 13 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ QCS8550 (Proxy) 0.067 ms 0 - 8 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS8450 0.1 ms 0 - 13 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-9075 0.156 ms 0 - 3 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-X7181 0.044 ms 7 - 7 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ Q-8750 0.049 ms 0 - 7 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Mobile 0.049 ms 0 - 7 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Gen 5 Mobile 0.051 ms 0 - 7 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® X2 Elite 0.123 ms 1 - 1 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® X Elite 0.138 ms 1 - 1 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 3 Mobile 0.057 ms 0 - 7 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 1 Mobile 0.096 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8275 0.309 ms 0 - 8 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8550 (Proxy) 0.074 ms 0 - 1 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8775P 0.205 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8650P 0.205 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8255P 0.205 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® QCS8450 0.096 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-9075 0.141 ms 0 - 2 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-X7181 0.138 ms 1 - 1 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ Q-8750 0.046 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA7255P 0.309 ms 0 - 8 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8295P 0.332 ms 0 - 6 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Mobile 0.046 ms 0 - 9 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 0.053 ms 0 - 8 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® X2 Elite 5.565 ms 10 - 10 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® X Elite 9.447 ms 142 - 142 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile 6.761 ms 9 - 19 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 1 Mobile 12.001 ms 9 - 21 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ QCS8550 (Proxy) 8.991 ms 0 - 3 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS8450 12.001 ms 9 - 21 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-9075 10.3 ms 8 - 12 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-X7181 9.447 ms 142 - 142 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ Q-8750 5.422 ms 9 - 22 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Mobile 5.422 ms 9 - 22 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Gen 5 Mobile 4.882 ms 10 - 21 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® X2 Elite 5.984 ms 3 - 3 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® X Elite 9.402 ms 3 - 3 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 3 Mobile 6.565 ms 2 - 11 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 1 Mobile 12.073 ms 2 - 12 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8275 23.475 ms 3 - 11 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8550 (Proxy) 8.88 ms 3 - 4 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8775P 10.672 ms 3 - 12 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8650P 10.672 ms 3 - 12 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8255P 10.672 ms 3 - 12 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® QCS8450 12.073 ms 2 - 12 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-9075 10.136 ms 3 - 7 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-X7181 9.402 ms 3 - 3 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ Q-8750 5.422 ms 3 - 15 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA7255P 23.475 ms 3 - 11 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8295P 13.057 ms 0 - 6 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Mobile 5.422 ms 3 - 15 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 4.88 ms 1 - 9 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® X2 Elite 0.088 ms 1 - 1 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® X Elite 0.141 ms 6 - 6 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile 0.143 ms 0 - 6 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 1 Mobile 0.237 ms 0 - 13 MB NPU
joiner PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ QCS8550 (Proxy) 0.193 ms 0 - 9 MB NPU
joiner PRECOMPILED_QNN_ONNX float Qualcomm® QCS8450 0.237 ms 0 - 13 MB NPU
joiner PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-9075 0.269 ms 0 - 3 MB NPU
joiner PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ IQ-X7181 0.141 ms 6 - 6 MB NPU
joiner PRECOMPILED_QNN_ONNX float Qualcomm® Dragonwing™ Q-8750 0.141 ms 0 - 11 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Mobile 0.141 ms 0 - 11 MB NPU
joiner PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Gen 5 Mobile 0.135 ms 0 - 7 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® X2 Elite 0.175 ms 0 - 0 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® X Elite 0.251 ms 0 - 0 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 3 Mobile 0.155 ms 0 - 8 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 1 Mobile 0.229 ms 0 - 9 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8275 0.475 ms 0 - 8 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ QCS8550 (Proxy) 0.181 ms 0 - 1 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® SA8775P 0.323 ms 0 - 10 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® SA8650P 0.323 ms 0 - 10 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® SA8255P 0.323 ms 0 - 10 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® QCS8450 0.229 ms 0 - 9 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-9075 0.236 ms 0 - 2 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ IQ-X7181 0.251 ms 0 - 0 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® Dragonwing™ Q-8750 0.13 ms 0 - 9 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® SA7255P 0.475 ms 0 - 8 MB NPU
joiner QNN_CONTEXT_BINARY float Qualcomm® SA8295P 0.455 ms 0 - 6 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Mobile 0.13 ms 0 - 9 MB NPU
joiner QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 0.13 ms 0 - 8 MB NPU

License

  • The license for the original implementation of Zipformer can be found here.

References

Community

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Automatic Speech Recognition
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for qualcomm/Zipformer