Hugging Face's logo

qualcomm
/
Whisper-Large-V3-Turbo

Automatic Speech Recognition
PyTorch
foundation
android
Model card Files
xet
 Community

Whisper-Large-V3-Turbo: Optimized for Qualcomm Devices

Whisper large-v3-turbo is a finetuned version of a pruned Whisper large-v3. In other words, it's the exact same model, except that the number of decoding layers have reduced from 32 to 4. As a result, the model is way faster, at the expense of a minor quality degradation. 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-Large-V3-Turbo 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 qualcomm_qcs8550_proxy QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_qcs9075 QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_snapdragon_8_elite_for_galaxy QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_snapdragon_8_elite_gen5 QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_snapdragon_8gen3 QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_snapdragon_x2_elite QAIRT 2.42, ONNX Runtime 1.24.3 Download
PRECOMPILED_QNN_ONNX float qualcomm_snapdragon_x_elite QAIRT 2.42, ONNX Runtime 1.24.3 Download
QNN_CONTEXT_BINARY float qualcomm_qcs8450_proxy QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_qcs8550_proxy QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_qcs9075 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_sa8775p QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_snapdragon_8_elite_for_galaxy QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_snapdragon_8_elite_gen5 QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_snapdragon_8gen3 QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_snapdragon_x2_elite QAIRT 2.43 Download
QNN_CONTEXT_BINARY float qualcomm_snapdragon_x_elite QAIRT 2.43 Download

For more device-specific assets and performance metrics, visit Whisper-Large-V3-Turbo 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-Large-V3-Turbo on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.speech_recognition

Model Stats:

  • Model checkpoint: openai/whisper-large-v3-turbo
  • Input resolution: 128x3000 (30 seconds audio)
  • Max decoded sequence length: 200 tokens

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 6.141 ms 42 - 52 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® X2 Elite 4.546 ms 399 - 399 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® X Elite 8.301 ms 399 - 399 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile 7.703 ms 44 - 55 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS8550 (Proxy) 9.71 ms 34 - 36 MB NPU
decoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS9075 10.284 ms 33 - 69 MB NPU
decoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite For Galaxy Mobile 6.631 ms 20 - 30 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 6.163 ms 33 - 43 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® X2 Elite 5.099 ms 33 - 33 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® X Elite 8.106 ms 33 - 33 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 3 Mobile 7.805 ms 33 - 41 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® QCS8275 (Proxy) 15.264 ms 33 - 40 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® QCS8550 (Proxy) 9.658 ms 33 - 38 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8775P 10.554 ms 33 - 40 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® QCS9075 10.323 ms 33 - 72 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® QCS8450 (Proxy) 16.33 ms 33 - 42 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA7255P 15.264 ms 33 - 40 MB NPU
decoder QNN_CONTEXT_BINARY float Qualcomm® SA8295P 11.891 ms 33 - 38 MB NPU
decoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite For Galaxy Mobile 6.616 ms 4 - 17 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite Gen 5 Mobile 249.498 ms 62 - 70 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® X2 Elite 250.303 ms 1276 - 1276 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® X Elite 585.925 ms 1274 - 1274 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Gen 3 Mobile 422.988 ms 64 - 75 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS8550 (Proxy) 577.46 ms 0 - 1282 MB NPU
encoder PRECOMPILED_QNN_ONNX float Qualcomm® QCS9075 700.978 ms 33 - 36 MB NPU
encoder PRECOMPILED_QNN_ONNX float Snapdragon® 8 Elite For Galaxy Mobile 317.001 ms 64 - 76 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite Gen 5 Mobile 248.558 ms 7 - 16 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® X2 Elite 247.321 ms 1 - 1 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® X Elite 580.939 ms 1 - 1 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Gen 3 Mobile 416.445 ms 1 - 8 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® QCS8275 (Proxy) 2174.853 ms 1 - 7 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® QCS8550 (Proxy) 579.071 ms 1 - 3 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8775P 707.156 ms 1 - 7 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® QCS9075 699.773 ms 1 - 32 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® QCS8450 (Proxy) 1312.236 ms 1 - 15 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA7255P 2174.853 ms 1 - 7 MB NPU
encoder QNN_CONTEXT_BINARY float Qualcomm® SA8295P 864.952 ms 1 - 10 MB NPU
encoder QNN_CONTEXT_BINARY float Snapdragon® 8 Elite For Galaxy Mobile 306.785 ms 1 - 14 MB NPU

License

  • The license for the original implementation of Whisper-Large-V3-Turbo can be found here.

References

Community

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Automatic Speech Recognition
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support