Whisper-Base-En: Optimized for Mobile Deployment
Automatic speech recognition (ASR) model for English transcription as well as translation
OpenAI’s Whisper ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text. 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 mean decoded length specified below.
This model is an implementation of Whisper-Base-En found here.
This repository provides scripts to run Whisper-Base-En on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Speech recognition
- Model Stats:
- Model checkpoint: base.en
- Input resolution: 80x3000 (30 seconds audio)
- Mean decoded sequence length: 112 tokens
- Number of parameters (WhisperEncoder): 23.7M
- Model size (WhisperEncoder): 90.6 MB
- Number of parameters (WhisperDecoder): 48.6M
- Model size (WhisperDecoder): 186 MB
Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |
---|---|---|---|---|---|---|---|---|
WhisperDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 37.916 ms | 5 - 43 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 4.223 ms | 20 - 63 MB | FP16 | NPU | Whisper-Base-En.so |
WhisperDecoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 15.843 ms | 0 - 116 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 32.58 ms | 4 - 82 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 3.296 ms | 20 - 75 MB | FP16 | NPU | Whisper-Base-En.so |
WhisperDecoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 13.657 ms | 0 - 222 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 27.503 ms | 4 - 76 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 2.646 ms | 9 - 59 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 12.074 ms | 6 - 203 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperDecoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 38.524 ms | 6 - 42 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 4.167 ms | 19 - 20 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | SA7255P ADP | SA7255P | TFLITE | 67.034 ms | 4 - 76 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | SA7255P ADP | SA7255P | QNN | 26.93 ms | 9 - 19 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 38.064 ms | 6 - 42 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 4.245 ms | 19 - 20 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | SA8295P ADP | SA8295P | TFLITE | 40.976 ms | 2 - 67 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | SA8295P ADP | SA8295P | QNN | 5.535 ms | 18 - 24 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 38.218 ms | 6 - 43 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 4.172 ms | 20 - 22 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | SA8775P ADP | SA8775P | TFLITE | 38.342 ms | 5 - 77 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | SA8775P ADP | SA8775P | QNN | 5.638 ms | 18 - 28 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 42.473 ms | 3 - 75 MB | FP16 | NPU | Whisper-Base-En.tflite |
WhisperDecoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 5.146 ms | 20 - 71 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 3.907 ms | 20 - 20 MB | FP16 | NPU | Use Export Script |
WhisperDecoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 14.267 ms | 107 - 107 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 196.633 ms | 35 - 110 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 266.434 ms | 0 - 77 MB | FP16 | NPU | Whisper-Base-En.so |
WhisperEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 300.379 ms | 91 - 94 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 157.913 ms | 37 - 77 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 186.485 ms | 1 - 287 MB | FP16 | NPU | Whisper-Base-En.so |
WhisperEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 211.549 ms | 94 - 982 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 140.336 ms | 37 - 60 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 152.984 ms | 0 - 304 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 183.176 ms | 96 - 640 MB | FP16 | NPU | Whisper-Base-En.onnx |
WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 196.378 ms | 34 - 110 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 199.737 ms | 0 - 21 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | SA7255P ADP | SA7255P | TFLITE | 1160.066 ms | 32 - 56 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | SA7255P ADP | SA7255P | QNN | 935.326 ms | 1 - 11 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 200.355 ms | 34 - 123 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 217.297 ms | 1 - 2 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | SA8295P ADP | SA8295P | TFLITE | 204.198 ms | 38 - 66 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | SA8295P ADP | SA8295P | QNN | 219.469 ms | 1 - 7 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 200.209 ms | 0 - 78 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 213.794 ms | 0 - 11 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | SA8775P ADP | SA8775P | TFLITE | 367.334 ms | 37 - 64 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | SA8775P ADP | SA8775P | QNN | 195.132 ms | 0 - 6 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 284.098 ms | 39 - 83 MB | FP16 | GPU | Whisper-Base-En.tflite |
WhisperEncoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 283.131 ms | 0 - 297 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 160.208 ms | 0 - 0 MB | FP16 | NPU | Use Export Script |
WhisperEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 297.781 ms | 133 - 133 MB | FP16 | NPU | Whisper-Base-En.onnx |
Installation
This model can be installed as a Python package via pip.
pip install "qai-hub-models[whisper_base_en]"
Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
Sign-in to Qualcomm® AI Hub 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.
qai-hub configure --api_token API_TOKEN
Navigate to 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.
python -m qai_hub_models.models.whisper_base_en.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_base_en.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.
python -m qai_hub_models.models.whisper_base_en.export
Profiling Results
------------------------------------------------------------
WhisperDecoder
Device : Samsung Galaxy S23 (13)
Runtime : TFLITE
Estimated inference time (ms) : 37.9
Estimated peak memory usage (MB): [5, 43]
Total # Ops : 983
Compute Unit(s) : NPU (983 ops)
------------------------------------------------------------
WhisperEncoder
Device : Samsung Galaxy S23 (13)
Runtime : TFLITE
Estimated inference time (ms) : 196.6
Estimated peak memory usage (MB): [35, 110]
Total # Ops : 419
Compute Unit(s) : GPU (408 ops) CPU (11 ops)
How does this work?
This export script leverages Qualcomm® AI Hub 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.
import torch
import qai_hub as hub
from qai_hub_models.models.whisper_base_en import Model
# Load the model
model = Model.from_pretrained()
decoder_model = model.decoder
encoder_model = model.encoder
# Device
device = hub.Device("Samsung Galaxy S23")
# Trace model
decoder_input_shape = decoder_model.get_input_spec()
decoder_sample_inputs = decoder_model.sample_inputs()
traced_decoder_model = torch.jit.trace(decoder_model, [torch.tensor(data[0]) for _, data in decoder_sample_inputs.items()])
# Compile model on a specific device
decoder_compile_job = hub.submit_compile_job(
model=traced_decoder_model ,
device=device,
input_specs=decoder_model.get_input_spec(),
)
# Get target model to run on-device
decoder_target_model = decoder_compile_job.get_target_model()
# Trace model
encoder_input_shape = encoder_model.get_input_spec()
encoder_sample_inputs = encoder_model.sample_inputs()
traced_encoder_model = torch.jit.trace(encoder_model, [torch.tensor(data[0]) for _, data in encoder_sample_inputs.items()])
# Compile model on a specific device
encoder_compile_job = hub.submit_compile_job(
model=traced_encoder_model ,
device=device,
input_specs=encoder_model.get_input_spec(),
)
# Get target model to run on-device
encoder_target_model = encoder_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.
decoder_profile_job = hub.submit_profile_job(
model=decoder_target_model,
device=device,
)
encoder_profile_job = hub.submit_profile_job(
model=encoder_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.
decoder_input_data = decoder_model.sample_inputs()
decoder_inference_job = hub.submit_inference_job(
model=decoder_target_model,
device=device,
inputs=decoder_input_data,
)
decoder_inference_job.download_output_data()
encoder_input_data = encoder_model.sample_inputs()
encoder_inference_job = hub.submit_inference_job(
model=encoder_target_model,
device=device,
inputs=encoder_input_data,
)
encoder_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. Sign up for access.
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tflite
export): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.so
export ): This sample app provides instructions on how to use the.so
shared library in an Android application.
View on Qualcomm® AI Hub
Get more details on Whisper-Base-En's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of Whisper-Base-En can be found here.
- The license for the compiled assets for on-device deployment 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.