--- 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_base/web-assets/model_demo.png) # 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](https://github.com/huggingface/transformers/tree/v4.42.3/src/transformers/models/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/whisper_base) 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 | |---|---|---|---|---| | PRECOMPILED_QNN_ONNX | float | qualcomm_qcs8550_proxy | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_qcs8550_proxy.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_qcs9075 | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_qcs9075.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_snapdragon_8_elite_for_galaxy | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_for_galaxy.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_snapdragon_8_elite_gen5 | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_gen5.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_snapdragon_8gen3 | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_snapdragon_8gen3.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_snapdragon_x2_elite | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_snapdragon_x2_elite.zip) | PRECOMPILED_QNN_ONNX | float | qualcomm_snapdragon_x_elite | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-precompiled_qnn_onnx-float-qualcomm_snapdragon_x_elite.zip) | QNN_CONTEXT_BINARY | float | qualcomm_qcs8450_proxy | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_qcs8450_proxy.zip) | QNN_CONTEXT_BINARY | float | qualcomm_qcs8550_proxy | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_qcs8550_proxy.zip) | QNN_CONTEXT_BINARY | float | qualcomm_qcs9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_qcs9075.zip) | QNN_CONTEXT_BINARY | float | qualcomm_sa7255p | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_sa7255p.zip) | QNN_CONTEXT_BINARY | float | qualcomm_sa8295p | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_sa8295p.zip) | QNN_CONTEXT_BINARY | float | qualcomm_sa8775p | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_sa8775p.zip) | QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8_elite_for_galaxy | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_snapdragon_8_elite_for_galaxy.zip) | QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8_elite_gen5 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_snapdragon_8_elite_gen5.zip) | QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8gen3 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_snapdragon_8gen3.zip) | QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_x2_elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_snapdragon_x2_elite.zip) | QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_x_elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/whisper_base/releases/v0.50.2/whisper_base-qnn_context_binary-float-qualcomm_snapdragon_x_elite.zip) For more device-specific assets and performance metrics, visit **[Whisper-Base on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/whisper_base)**. ### 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/whisper_base) 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/whisper_base) 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 (encoder): 23.7M - Model size (encoder) (float): 90.7 MB - Number of parameters (decoder): 48.9M - Model size (decoder) (float): 187 MB ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.48 ms | 20 - 31 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 1.842 ms | 126 - 126 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 3.388 ms | 125 - 125 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 3.268 ms | 16 - 23 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 4.005 ms | 20 - 22 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 4.626 ms | 20 - 43 MB | NPU | decoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.676 ms | 19 - 26 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.473 ms | 20 - 30 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 2.333 ms | 20 - 20 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 3.367 ms | 20 - 20 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 3.132 ms | 0 - 8 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 6.572 ms | 20 - 28 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 3.869 ms | 17 - 19 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 4.854 ms | 11 - 19 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 4.666 ms | 20 - 44 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 5.12 ms | 20 - 29 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 6.572 ms | 20 - 28 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 5.436 ms | 18 - 23 MB | NPU | decoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.65 ms | 0 - 14 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.994 ms | 39 - 49 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 15.659 ms | 65 - 65 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 37.403 ms | 67 - 67 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 26.845 ms | 42 - 48 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 36.375 ms | 8 - 56 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 44.961 ms | 39 - 42 MB | NPU | encoder | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.531 ms | 40 - 49 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.55 ms | 0 - 10 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 15.506 ms | 0 - 0 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 37.227 ms | 0 - 0 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 27.135 ms | 1 - 8 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 119.078 ms | 1 - 10 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 36.57 ms | 1 - 2 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 43.732 ms | 1 - 10 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 44.523 ms | 0 - 20 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 97.637 ms | 1 - 10 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 119.078 ms | 1 - 10 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 74.047 ms | 0 - 5 MB | NPU | encoder | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.336 ms | 1 - 11 MB | NPU ## License * The license for the original implementation of Whisper-Base 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).