--- library_name: pytorch license: other tags: - foundation - android pipeline_tag: automatic-speech-recognition --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_small/web-assets/model_demo.png) # Whisper-Small: Optimized for Mobile Deployment ## Transformer-based automatic speech recognition (ASR) model for multilingual transcription and translation available on HuggingFace 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 model is an implementation of Whisper-Small found [here](https://github.com/huggingface/transformers/tree/v4.42.3/src/transformers/models/whisper). This repository provides scripts to run Whisper-Small on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/whisper_small). ### Model Details - **Model Type:** Model_use_case.speech_recognition - **Model Stats:** - Model checkpoint: openai/whisper-small - Input resolution: 80x3000 (30 seconds audio) - Max decoded sequence length: 200 tokens - Number of parameters (HfWhisperEncoder): 102M - Model size (HfWhisperEncoder) (float): 391 MB - Number of parameters (HfWhisperDecoder): 139M - Model size (HfWhisperDecoder) (float): 533 MB | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | HfWhisperEncoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_CONTEXT_BINARY | 427.44 ms | 1 - 9 MB | NPU | Use Export Script | | HfWhisperEncoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_CONTEXT_BINARY | 279.879 ms | 1 - 20 MB | NPU | Use Export Script | | HfWhisperEncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_CONTEXT_BINARY | 120.06 ms | 1 - 4 MB | NPU | Use Export Script | | HfWhisperEncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | PRECOMPILED_QNN_ONNX | 135.925 ms | 0 - 259 MB | NPU | Use Export Script | | HfWhisperEncoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_CONTEXT_BINARY | 623.288 ms | 1 - 9 MB | NPU | Use Export Script | | HfWhisperEncoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_CONTEXT_BINARY | 427.44 ms | 1 - 9 MB | NPU | Use Export Script | | HfWhisperEncoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_CONTEXT_BINARY | 248.677 ms | 1 - 15 MB | NPU | Use Export Script | | HfWhisperEncoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_CONTEXT_BINARY | 623.288 ms | 1 - 9 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_CONTEXT_BINARY | 87.154 ms | 0 - 20 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | PRECOMPILED_QNN_ONNX | 104.052 ms | 129 - 148 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_CONTEXT_BINARY | 59.805 ms | 0 - 16 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | PRECOMPILED_QNN_ONNX | 81.116 ms | 127 - 146 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_CONTEXT_BINARY | 45.589 ms | 1 - 11 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | PRECOMPILED_QNN_ONNX | 63.902 ms | 124 - 135 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_CONTEXT_BINARY | 120.619 ms | 0 - 0 MB | NPU | Use Export Script | | HfWhisperEncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | PRECOMPILED_QNN_ONNX | 130.765 ms | 226 - 226 MB | NPU | Use Export Script | | HfWhisperDecoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_CONTEXT_BINARY | 18.85 ms | 47 - 57 MB | NPU | Use Export Script | | HfWhisperDecoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_CONTEXT_BINARY | 18.187 ms | 60 - 77 MB | NPU | Use Export Script | | HfWhisperDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_CONTEXT_BINARY | 11.815 ms | 60 - 62 MB | NPU | Use Export Script | | HfWhisperDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | PRECOMPILED_QNN_ONNX | 12.719 ms | 58 - 60 MB | NPU | Use Export Script | | HfWhisperDecoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_CONTEXT_BINARY | 13.261 ms | 55 - 64 MB | NPU | Use Export Script | | HfWhisperDecoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_CONTEXT_BINARY | 18.85 ms | 47 - 57 MB | NPU | Use Export Script | | HfWhisperDecoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_CONTEXT_BINARY | 14.62 ms | 51 - 64 MB | NPU | Use Export Script | | HfWhisperDecoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_CONTEXT_BINARY | 13.261 ms | 55 - 64 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_CONTEXT_BINARY | 9.439 ms | 60 - 78 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | PRECOMPILED_QNN_ONNX | 9.983 ms | 75 - 94 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_CONTEXT_BINARY | 8.174 ms | 19 - 36 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | PRECOMPILED_QNN_ONNX | 8.573 ms | 53 - 63 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_CONTEXT_BINARY | 7.278 ms | 60 - 71 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | PRECOMPILED_QNN_ONNX | 7.543 ms | 75 - 85 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_CONTEXT_BINARY | 9.939 ms | 60 - 60 MB | NPU | Use Export Script | | HfWhisperDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | PRECOMPILED_QNN_ONNX | 10.169 ms | 286 - 286 MB | NPU | Use Export Script | ## Installation Install the package via pip: ```bash # NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported. pip install "qai-hub-models[whisper-small]" ``` ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) with your Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`. With this API token, you can configure your client to run models on the cloud hosted devices. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://workbench.aihub.qualcomm.com/docs/) for more information. ## Demo off target The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input. ```bash python -m qai_hub_models.models.whisper_small.demo ``` The above demo runs a reference implementation of pre-processing, model inference, and post processing. **NOTE**: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above). ``` %run -m qai_hub_models.models.whisper_small.demo ``` ### Run model on a cloud-hosted device In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following: * Performance check on-device on a cloud-hosted device * Downloads compiled assets that can be deployed on-device for Android. * Accuracy check between PyTorch and on-device outputs. ```bash python -m qai_hub_models.models.whisper_small.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/whisper_small/qai_hub_models/models/Whisper-Small/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model on-device. Lets go through each step below in detail: Step 1: **Compile model for on-device deployment** To compile a PyTorch model for on-device deployment, we first trace the model in memory using the `jit.trace` and then call the `submit_compile_job` API. ```python import torch import qai_hub as hub from qai_hub_models.models.whisper_small import Model # Load the model torch_model = Model.from_pretrained() # Device device = hub.Device("Samsung Galaxy S25") # Trace model input_shape = torch_model.get_input_spec() sample_inputs = torch_model.sample_inputs() pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()]) # Compile model on a specific device compile_job = hub.submit_compile_job( model=pt_model, device=device, input_specs=torch_model.get_input_spec(), ) # Get target model to run on-device target_model = compile_job.get_target_model() ``` Step 2: **Performance profiling on cloud-hosted device** After compiling models from step 1. Models can be profiled model on-device using the `target_model`. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics. ```python profile_job = hub.submit_profile_job( model=target_model, device=device, ) ``` Step 3: **Verify on-device accuracy** To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device. ```python input_data = torch_model.sample_inputs() inference_job = hub.submit_inference_job( model=target_model, device=device, inputs=input_data, ) on_device_output = inference_job.download_output_data() ``` With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output. **Note**: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup). ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on Whisper-Small's performance across various devices [here](https://aihub.qualcomm.com/models/whisper_small). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of Whisper-Small can be found [here](https://github.com/huggingface/transformers/blob/v4.42.3/LICENSE). ## References * [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) * [Source Model Implementation](https://github.com/huggingface/transformers/tree/v4.42.3/src/transformers/models/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).