---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: automatic-speech-recognition

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/web-assets/model_demo.png)

# 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](https://github.com/huggingface/distil-whisper).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) 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 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.51.0/distil_whisper-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.51.0/distil_whisper-qnn_dlc-float.zip)
| TFLITE | float | Universal |  | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/distil_whisper/releases/v0.51.0/distil_whisper-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[Distil-Whisper on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/distil_whisper)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/distil_whisper) 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® 8 Elite Gen 5 Mobile | 5.551 ms | 52 - 376 MB | NPU
| decoder | ONNX | float | Snapdragon® X2 Elite | 5.092 ms | 178 - 178 MB | NPU
| decoder | ONNX | float | Snapdragon® X Elite | 11.298 ms | 178 - 178 MB | NPU
| decoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.634 ms | 52 - 427 MB | NPU
| decoder | ONNX | float | Qualcomm® QCS8550 (Proxy) | 11.745 ms | 0 - 183 MB | NPU
| decoder | ONNX | float | Qualcomm® QCS9075 | 13.145 ms | 40 - 82 MB | NPU
| decoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.208 ms | 16 - 473 MB | NPU
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.682 ms | 32 - 564 MB | NPU
| decoder | QNN_DLC | float | Snapdragon® X2 Elite | 5.897 ms | 40 - 40 MB | NPU
| decoder | QNN_DLC | float | Snapdragon® X Elite | 11.1 ms | 40 - 40 MB | NPU
| decoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 8.72 ms | 0 - 444 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 19.412 ms | 16 - 357 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 11.692 ms | 40 - 46 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® SA8775P | 12.852 ms | 30 - 371 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® QCS9075 | 19.477 ms | 40 - 86 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 17.997 ms | 36 - 336 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® SA7255P | 19.412 ms | 16 - 357 MB | NPU
| decoder | QNN_DLC | float | Qualcomm® SA8295P | 13.897 ms | 16 - 265 MB | NPU
| decoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.214 ms | 5 - 548 MB | NPU
| decoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.839 ms | 4 - 570 MB | NPU
| decoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 8.704 ms | 0 - 740 MB | NPU
| decoder | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 19.254 ms | 5 - 538 MB | NPU
| decoder | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 11.542 ms | 5 - 7 MB | NPU
| decoder | TFLITE | float | Qualcomm® SA8775P | 13.059 ms | 5 - 537 MB | NPU
| decoder | TFLITE | float | Qualcomm® QCS9075 | 16.359 ms | 0 - 265 MB | NPU
| decoder | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 18.33 ms | 5 - 466 MB | NPU
| decoder | TFLITE | float | Qualcomm® SA7255P | 19.254 ms | 5 - 538 MB | NPU
| decoder | TFLITE | float | Qualcomm® SA8295P | 14.216 ms | 5 - 297 MB | NPU
| decoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 7.215 ms | 4 - 574 MB | NPU
| encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 51.408 ms | 79 - 838 MB | NPU
| encoder | ONNX | float | Snapdragon® X2 Elite | 51.169 ms | 183 - 183 MB | NPU
| encoder | ONNX | float | Snapdragon® X Elite | 123.64 ms | 182 - 182 MB | NPU
| encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 82.241 ms | 0 - 1165 MB | NPU
| encoder | ONNX | float | Qualcomm® QCS8550 (Proxy) | 117.803 ms | 2 - 199 MB | NPU
| encoder | ONNX | float | Qualcomm® QCS9075 | 150.38 ms | 77 - 80 MB | NPU
| encoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 60.531 ms | 77 - 731 MB | NPU
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 60.083 ms | 1 - 712 MB | NPU
| encoder | QNN_DLC | float | Snapdragon® X2 Elite | 59.899 ms | 1 - 1 MB | NPU
| encoder | QNN_DLC | float | Snapdragon® X Elite | 139.121 ms | 1 - 1 MB | NPU
| encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 97.09 ms | 0 - 964 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 437.445 ms | 1 - 696 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 135.417 ms | 1 - 920 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® SA8775P | 153.18 ms | 1 - 688 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® QCS9075 | 170.284 ms | 1 - 39 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 268.98 ms | 0 - 826 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® SA7255P | 437.445 ms | 1 - 696 MB | NPU
| encoder | QNN_DLC | float | Qualcomm® SA8295P | 192.899 ms | 1 - 611 MB | NPU
| encoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.51 ms | 1 - 691 MB | NPU
| encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 417.61 ms | 40 - 81 MB | GPU
| encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 480.056 ms | 0 - 145 MB | GPU
| encoder | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 3135.88 ms | 32 - 76 MB | GPU
| encoder | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 660.287 ms | 0 - 227 MB | GPU
| encoder | TFLITE | float | Qualcomm® SA8775P | 1324.546 ms | 18 - 61 MB | GPU
| encoder | TFLITE | float | Qualcomm® QCS9075 | 1274.677 ms | 0 - 40 MB | GPU
| encoder | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 845.162 ms | 40 - 195 MB | GPU
| encoder | TFLITE | float | Qualcomm® SA7255P | 3135.88 ms | 32 - 76 MB | GPU
| encoder | TFLITE | float | Qualcomm® SA8295P | 669.29 ms | 38 - 81 MB | GPU
| encoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 408.665 ms | 42 - 81 MB | GPU

## License
* The license for the original implementation of Distil-Whisper can be found
  [here](https://github.com/huggingface/distil-whisper/blob/main/LICENSE).

## References
* [Distil-Whisper - Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://arxiv.org/abs/2311.00430)
* [Source Model Implementation](https://github.com/huggingface/distil-whisper)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).