Hugging Face's logo

qualcomm
/
Whisper-Large-V3-Turbo-Quantized

Automatic Speech Recognition
PyTorch
foundation
android
Model card Files
xet
 Community

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

We have applied w8a16 quantization to significantly enhance performance and efficiency. Whisper large-v3-turbo is a finetuned version of a pruned Whisper large-v3, with the number of decoding layers reduced from 32 to 4. 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-Quantized 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.

Deploying Whisper-Large-V3-Turbo-Quantized on-device

This model is compatible with the Qualcomm Voice AI SDK. Download the SDK from the Qualcomm Package Manager to deploy this model on-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 w8a16 Snapdragon® X2 Elite QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Snapdragon® X Elite QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Gen 3 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-8275 QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ Q-6690 QAIRT 2.45 Download
PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-9075 QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 7 Gen 4 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45, ONNX Runtime 1.27.1 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® X2 Elite QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® X Elite QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Gen 3 Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-8275 QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8775P QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ Q-6690 QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-9075 QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Qualcomm® SA7255P QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® 7 Gen 4 Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® X2 Elite QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® X Elite QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® 8 Gen 3 Mobile QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-8275 QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® SA8775P QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® Dragonwing™ Q-6690 QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-9075 QAIRT 2.45 Download
VOICE_AI w8a16 Qualcomm® SA7255P QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® 7 Gen 4 Mobile QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® 8 Elite Mobile QAIRT 2.45 Download
VOICE_AI w8a16 Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45 Download

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

Model Details

Model Type: Model_use_case.speech_recognition

Model Stats:

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

Performance Summary

Model Runtime Precision Chipset Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® X2 Elite 2.6 ms 17 - 17 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® X Elite 5.343 ms 267 - 267 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Gen 3 Mobile 4.478 ms 21 - 34 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Gen 1 Mobile 8.06 ms 21 - 35 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-8275 6.885 ms 13 - 33 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 6.009 ms 0 - 304 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® QCS8450 8.06 ms 21 - 35 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-9075 6.673 ms 12 - 31 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-X7181 5.343 ms 267 - 267 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ Q-7790 8.444 ms 16 - 25 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ Q-8750 3.62 ms 21 - 34 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Mobile 3.62 ms 21 - 34 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Gen 5 Mobile 3.292 ms 15 - 28 MB NPU
decoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 7 Gen 4 Mobile 8.444 ms 16 - 25 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® X2 Elite 3.066 ms 17 - 17 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® X Elite 5.644 ms 17 - 17 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Gen 3 Mobile 4.414 ms 17 - 24 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-8275 6.736 ms 16 - 37 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-8275 9.116 ms 17 - 26 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 5.786 ms 17 - 18 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8775P 6.607 ms 17 - 26 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8650P 6.607 ms 17 - 26 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8255P 6.607 ms 17 - 26 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-9075 6.479 ms 14 - 34 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-X7181 5.644 ms 17 - 17 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ Q-7790 8.407 ms 16 - 22 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ Q-8750 3.553 ms 15 - 24 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA7255P 9.116 ms 17 - 26 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Mobile 3.553 ms 15 - 24 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Gen 5 Mobile 3.29 ms 17 - 25 MB NPU
decoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 7 Gen 4 Mobile 8.407 ms 16 - 22 MB NPU
decoder VOICE_AI w8a16 Snapdragon® X2 Elite 3.167 ms 17 - 17 MB NPU
decoder VOICE_AI w8a16 Snapdragon® X Elite 5.578 ms 17 - 17 MB NPU
decoder VOICE_AI w8a16 Snapdragon® 8 Gen 3 Mobile 4.411 ms 16 - 24 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-8275 9.046 ms 17 - 26 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 5.85 ms 17 - 18 MB NPU
decoder VOICE_AI w8a16 Qualcomm® SA8775P 6.651 ms 17 - 26 MB NPU
decoder VOICE_AI w8a16 Qualcomm® SA8650P 6.651 ms 17 - 26 MB NPU
decoder VOICE_AI w8a16 Qualcomm® SA8255P 6.651 ms 17 - 26 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-9075 6.469 ms 14 - 34 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-X7181 5.578 ms 17 - 17 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ Q-7790 8.428 ms 16 - 22 MB NPU
decoder VOICE_AI w8a16 Qualcomm® Dragonwing™ Q-8750 3.574 ms 2 - 11 MB NPU
decoder VOICE_AI w8a16 Qualcomm® SA7255P 9.046 ms 17 - 26 MB NPU
decoder VOICE_AI w8a16 Snapdragon® 8 Elite Mobile 3.574 ms 2 - 11 MB NPU
decoder VOICE_AI w8a16 Snapdragon® 8 Elite Gen 5 Mobile 3.315 ms 17 - 27 MB NPU
decoder VOICE_AI w8a16 Snapdragon® 7 Gen 4 Mobile 8.428 ms 16 - 22 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® X2 Elite 690.795 ms 33 - 33 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® X Elite 1342.677 ms 742 - 742 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Gen 3 Mobile 1401.054 ms 2 - 12 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Gen 1 Mobile 1539.984 ms 2 - 15 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-8275 1345.497 ms 1 - 5 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 1764.402 ms 1 - 770 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® QCS8450 1539.984 ms 2 - 15 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-9075 1296.094 ms 1 - 4 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ IQ-X7181 1342.677 ms 742 - 742 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ Q-7790 2236.001 ms 1 - 12 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Qualcomm® Dragonwing™ Q-8750 1251.539 ms 1 - 12 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Mobile 1251.539 ms 1 - 12 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 8 Elite Gen 5 Mobile 1263.617 ms 1 - 13 MB NPU
encoder PRECOMPILED_QNN_ONNX w8a16 Snapdragon® 7 Gen 4 Mobile 2236.001 ms 1 - 12 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® X2 Elite 689.52 ms 1 - 1 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® X Elite 1343.43 ms 1 - 1 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Gen 3 Mobile 1349.614 ms 1 - 9 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-8275 1342.794 ms 0 - 18 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-8275 2359.354 ms 1 - 9 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 1665.476 ms 2 - 14 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8775P 1434.451 ms 1 - 11 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8650P 1434.451 ms 1 - 11 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA8255P 1434.451 ms 1 - 11 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-9075 1295.662 ms 3 - 20 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ IQ-X7181 1343.43 ms 1 - 1 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ Q-7790 2225.298 ms 0 - 6 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® Dragonwing™ Q-8750 1223.092 ms 1 - 15 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Qualcomm® SA7255P 2359.354 ms 1 - 9 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Mobile 1223.092 ms 1 - 15 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 8 Elite Gen 5 Mobile 1141.032 ms 0 - 9 MB NPU
encoder QNN_CONTEXT_BINARY w8a16 Snapdragon® 7 Gen 4 Mobile 2225.298 ms 0 - 6 MB NPU
encoder VOICE_AI w8a16 Snapdragon® X2 Elite 690.347 ms 1 - 1 MB NPU
encoder VOICE_AI w8a16 Snapdragon® X Elite 1368.602 ms 1 - 1 MB NPU
encoder VOICE_AI w8a16 Snapdragon® 8 Gen 3 Mobile 1343.602 ms 1 - 9 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-8275 2349.213 ms 1 - 9 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ QCS8550 (Proxy) 1644.938 ms 1 - 4 MB NPU
encoder VOICE_AI w8a16 Qualcomm® SA8775P 1435.608 ms 1 - 11 MB NPU
encoder VOICE_AI w8a16 Qualcomm® SA8650P 1435.608 ms 1 - 11 MB NPU
encoder VOICE_AI w8a16 Qualcomm® SA8255P 1435.608 ms 1 - 11 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-9075 1294.958 ms 1 - 18 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ IQ-X7181 1368.602 ms 1 - 1 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ Q-7790 2230.833 ms 0 - 6 MB NPU
encoder VOICE_AI w8a16 Qualcomm® Dragonwing™ Q-8750 1192.536 ms 0 - 9 MB NPU
encoder VOICE_AI w8a16 Qualcomm® SA7255P 2349.213 ms 1 - 9 MB NPU
encoder VOICE_AI w8a16 Snapdragon® 8 Elite Mobile 1192.536 ms 0 - 9 MB NPU
encoder VOICE_AI w8a16 Snapdragon® 8 Elite Gen 5 Mobile 1115.986 ms 0 - 8 MB NPU
encoder VOICE_AI w8a16 Snapdragon® 7 Gen 4 Mobile 2230.833 ms 0 - 6 MB NPU

License

  • The license for the original implementation of Whisper-Large-V3-Turbo-Quantized 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