Distil-Whisper: Optimized for Qualcomm Devices

Distil-Whisper Small English is a distilled version of Whisper Small, optimized for fast and efficient automatic speech recognition.

This is based on the implementation of Distil-Whisper 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 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal		Download

For more device-specific assets and performance metrics, visit Distil-Whisper 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 Distil-Whisper on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.speech_recognition

Model Stats:

Model checkpoint: distil-whisper/distil-small.en
Input resolution: 80x3000 (30 seconds audio)
Max decoded sequence length: 200 tokens
Number of parameters (encoder): 166M
Model size (encoder) (float): 332 MB
Number of parameters (decoder): 211M
Model size (decoder) (float): 450MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
decoder	ONNX	float	Snapdragon® X2 Elite	5.275 ms	40 - 40 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	11.386 ms	180 - 180 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	8.894 ms	1 - 488 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	18.163 ms	50 - 396 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.757 ms	0 - 182 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	18.163 ms	50 - 396 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	12.828 ms	40 - 82 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	11.386 ms	180 - 180 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.27 ms	15 - 569 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Mobile	7.27 ms	15 - 569 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.746 ms	16 - 558 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	6.004 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	11.018 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.645 ms	0 - 603 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	18.094 ms	36 - 338 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	19.353 ms	29 - 527 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.461 ms	40 - 42 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8775P	12.778 ms	29 - 521 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8650P	12.778 ms	29 - 521 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8255P	12.778 ms	29 - 521 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	18.094 ms	36 - 338 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	15.152 ms	40 - 86 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	11.018 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	7.299 ms	0 - 543 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	19.353 ms	29 - 527 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	14.074 ms	17 - 258 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	7.299 ms	0 - 543 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.933 ms	1 - 503 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	8.63 ms	0 - 742 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	18.13 ms	5 - 467 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	19.233 ms	5 - 538 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.586 ms	5 - 8 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	12.982 ms	5 - 538 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	12.982 ms	5 - 538 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	12.982 ms	5 - 538 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	18.13 ms	5 - 467 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	12.802 ms	0 - 265 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	7.218 ms	3 - 572 MB	NPU
decoder	TFLITE	float	Qualcomm® SA7255P	19.233 ms	5 - 538 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	13.947 ms	5 - 297 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Mobile	7.218 ms	3 - 572 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	5.829 ms	4 - 575 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	60.272 ms	132 - 132 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	133.106 ms	237 - 237 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	97.214 ms	80 - 1156 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	269.48 ms	76 - 1014 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	130.192 ms	1 - 260 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	269.48 ms	76 - 1014 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	154.216 ms	80 - 83 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	133.106 ms	237 - 237 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	71.71 ms	42 - 747 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Mobile	71.71 ms	42 - 747 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	58.213 ms	77 - 806 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	60.548 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	139.211 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	97.057 ms	0 - 962 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	270.617 ms	0 - 823 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	438.216 ms	1 - 696 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	135.143 ms	1 - 28 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8775P	153.995 ms	1 - 686 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8650P	153.995 ms	1 - 686 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8255P	153.995 ms	1 - 686 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	270.617 ms	0 - 823 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	153.371 ms	1 - 39 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	139.211 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	71.443 ms	1 - 692 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	438.216 ms	1 - 696 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	193.086 ms	1 - 610 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	71.443 ms	1 - 692 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	57.724 ms	0 - 717 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	478.573 ms	42 - 187 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	865.917 ms	41 - 195 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	3107.385 ms	23 - 67 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	655.953 ms	0 - 310 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8775P	1324.065 ms	33 - 77 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8650P	1324.065 ms	33 - 77 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8255P	1324.065 ms	33 - 77 MB	GPU
encoder	TFLITE	float	Qualcomm® QCS8450	865.917 ms	41 - 195 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	1268.009 ms	0 - 40 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	406.823 ms	40 - 79 MB	GPU
encoder	TFLITE	float	Qualcomm® SA7255P	3107.385 ms	23 - 67 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8295P	670.254 ms	40 - 84 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Elite Mobile	406.823 ms	40 - 79 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	361.289 ms	42 - 81 MB	GPU

License

The license for the original implementation of Distil-Whisper 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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/Distil-Whisper

Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling

Paper • 2311.00430 • Published Nov 1, 2023 • 56