Whisper-Base: Optimized for Qualcomm Devices
HuggingFace Whisper-Small ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text. This model is based on the transformer architecture and has been optimized for edge inference by replacing Multi-Head Attention (MHA) with Single-Head Attention (SHA) and 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 audio clips up to 30 seconds long. Time to the first token is the encoder's latency, while time to each additional token is decoder's latency, where we assume a max decoded length specified below.
This is based on the implementation of Whisper-Base 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® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | Download |
For more device-specific assets and performance metrics, visit Whisper-Base 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 Whisper-Base on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.speech_recognition
Model Stats:
- Model checkpoint: openai/whisper-base
- Input resolution: 80x3000 (30 seconds audio)
- Max decoded sequence length: 200 tokens
- Number of parameters (HfWhisperEncoder): 23.7M
- Model size (HfWhisperEncoder) (float): 90.7 MB
- Number of parameters (HfWhisperDecoder): 48.9M
- Model size (HfWhisperDecoder) (float): 187 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 3.418 ms | 125 - 125 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 3.275 ms | 3 - 9 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 4.098 ms | 20 - 22 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 4.646 ms | 20 - 43 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.667 ms | 16 - 23 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.487 ms | 20 - 31 MB | NPU |
| HfWhisperDecoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 1.959 ms | 126 - 126 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 3.426 ms | 20 - 20 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 3.097 ms | 16 - 24 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 6.467 ms | 20 - 28 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 3.93 ms | 19 - 20 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 11.273 ms | 20 - 28 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 4.658 ms | 20 - 44 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 5.124 ms | 20 - 31 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 6.467 ms | 20 - 28 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 5.455 ms | 20 - 25 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.64 ms | 0 - 9 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.475 ms | 20 - 30 MB | NPU |
| HfWhisperDecoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 2.23 ms | 20 - 20 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 37.258 ms | 66 - 66 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 27.146 ms | 39 - 46 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 36.577 ms | 3 - 52 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 44.757 ms | 39 - 42 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.511 ms | 38 - 45 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.944 ms | 38 - 48 MB | NPU |
| HfWhisperEncoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 15.497 ms | 65 - 65 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 37.271 ms | 0 - 0 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 27.1 ms | 1 - 8 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 118.922 ms | 1 - 9 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 36.764 ms | 1 - 2 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 197.838 ms | 1 - 10 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 44.23 ms | 0 - 20 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 97.613 ms | 1 - 11 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 118.922 ms | 1 - 9 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 73.938 ms | 0 - 6 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.631 ms | 0 - 14 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.547 ms | 1 - 10 MB | NPU |
| HfWhisperEncoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 15.759 ms | 0 - 0 MB | NPU |
License
- The license for the original implementation of Whisper-Base 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.