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 2.50, ONNX Runtime 1.30.0	Download
QNN_DLC	float	Universal	QAIRT 2.50	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:

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

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	5.589 ms	10 - 354 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.321 ms	15 - 369 MB	NPU
decoder	ONNX	float	Snapdragon® X2 Elite	5.066 ms	40 - 40 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	11.242 ms	179 - 179 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	8.847 ms	23 - 497 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	14.093 ms	50 - 389 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	12.997 ms	40 - 84 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.697 ms	0 - 181 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	14.093 ms	50 - 389 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	12.811 ms	40 - 83 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	11.242 ms	179 - 179 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.321 ms	15 - 369 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	5.75 ms	1 - 331 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.371 ms	5 - 348 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	5.541 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	11.494 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.768 ms	0 - 428 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	13.627 ms	32 - 322 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	12.886 ms	40 - 87 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.496 ms	40 - 42 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	13.627 ms	32 - 322 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	12.615 ms	40 - 87 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	11.494 ms	40 - 40 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	7.371 ms	5 - 348 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	14.013 ms	5 - 201 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	5.866 ms	5 - 385 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	7.178 ms	4 - 370 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	8.717 ms	3 - 550 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	14.129 ms	5 - 453 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	12.735 ms	0 - 266 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.721 ms	5 - 8 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	13.055 ms	5 - 358 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	13.055 ms	5 - 358 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	14.129 ms	5 - 453 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	12.88 ms	0 - 265 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	7.178 ms	4 - 370 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	13.93 ms	5 - 260 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	53.931 ms	80 - 648 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	65.65 ms	62 - 608 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	55.186 ms	133 - 133 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	118.938 ms	236 - 236 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	90.041 ms	40 - 965 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	177.436 ms	76 - 965 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	148.425 ms	74 - 79 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	117.785 ms	0 - 258 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	177.436 ms	76 - 965 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	140.743 ms	80 - 84 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	118.938 ms	236 - 236 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	65.65 ms	62 - 608 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	53.71 ms	0 - 547 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	65.465 ms	0 - 526 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	55.499 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	118.403 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	87.105 ms	0 - 800 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	176.442 ms	0 - 765 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	145.694 ms	1 - 40 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	116.256 ms	4 - 10 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	176.442 ms	0 - 765 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	140.671 ms	1 - 39 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	118.403 ms	1 - 1 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	65.465 ms	0 - 526 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	174.649 ms	1 - 501 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	368.621 ms	42 - 83 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	416.444 ms	17 - 55 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	477.188 ms	14 - 157 MB	GPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	841.465 ms	40 - 194 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	2985.565 ms	0 - 39 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	655.81 ms	0 - 311 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8650P	1325.637 ms	38 - 83 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8255P	1325.637 ms	38 - 83 MB	GPU
encoder	TFLITE	float	Qualcomm® QCS8450	841.465 ms	40 - 194 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	1322.379 ms	0 - 39 MB	GPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	416.444 ms	17 - 55 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8295P	689.302 ms	38 - 82 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