Hugging Face's logo

qualcomm
/
Whisper-Base-En

Automatic Speech Recognition
PyTorch
LiteRT
ONNX
foundation
android
Model card Files Community
1

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
WhisperEncoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 TFLITE 207.168 ms 0 - 67 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 QNN 270.321 ms 0 - 355 MB FP16 NPU Whisper-Base-En.so
WhisperEncoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 ONNX 258.293 ms 53 - 564 MB FP16 NPU Whisper-Base-En.onnx
WhisperEncoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 TFLITE 159.338 ms 39 - 85 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 QNN 225.666 ms 0 - 1377 MB FP16 NPU Whisper-Base-En.so
WhisperEncoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 ONNX 165.997 ms 163 - 1665 MB FP16 NPU Whisper-Base-En.onnx
WhisperEncoderInf Snapdragon 8 Elite QRD Snapdragon® 8 Elite TFLITE 129.548 ms 39 - 67 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf Snapdragon 8 Elite QRD Snapdragon® 8 Elite ONNX 159.839 ms 41 - 1544 MB FP16 NPU Whisper-Base-En.onnx
WhisperEncoderInf SA7255P ADP SA7255P TFLITE 1153.482 ms 37 - 60 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf SA7255P ADP SA7255P QNN 1010.596 ms 1 - 10 MB FP16 NPU Use Export Script
WhisperEncoderInf SA8255 (Proxy) SA8255P Proxy TFLITE 204.974 ms 0 - 67 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf SA8255 (Proxy) SA8255P Proxy QNN 251.25 ms 1 - 3 MB FP16 NPU Use Export Script
WhisperEncoderInf SA8295P ADP SA8295P TFLITE 205.071 ms 38 - 69 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf SA8295P ADP SA8295P QNN 220.65 ms 1 - 17 MB FP16 NPU Use Export Script
WhisperEncoderInf SA8650 (Proxy) SA8650P Proxy TFLITE 233.593 ms 0 - 77 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf SA8650 (Proxy) SA8650P Proxy QNN 221.139 ms 1 - 2 MB FP16 NPU Use Export Script
WhisperEncoderInf SA8775P ADP SA8775P TFLITE 367.2 ms 38 - 62 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf SA8775P ADP SA8775P QNN 215.878 ms 0 - 9 MB FP16 NPU Use Export Script
WhisperEncoderInf QCS8275 (Proxy) QCS8275 Proxy TFLITE 1153.482 ms 37 - 60 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf QCS8275 (Proxy) QCS8275 Proxy QNN 1010.596 ms 1 - 10 MB FP16 NPU Use Export Script
WhisperEncoderInf QCS8550 (Proxy) QCS8550 Proxy TFLITE 201.097 ms 0 - 66 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf QCS8550 (Proxy) QCS8550 Proxy QNN 242.356 ms 1 - 4 MB FP16 NPU Use Export Script
WhisperEncoderInf QCS9075 (Proxy) QCS9075 Proxy TFLITE 367.2 ms 38 - 62 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf QCS9075 (Proxy) QCS9075 Proxy QNN 215.878 ms 0 - 9 MB FP16 NPU Use Export Script
WhisperEncoderInf QCS8450 (Proxy) QCS8450 Proxy TFLITE 264.328 ms 38 - 89 MB FP16 GPU Whisper-Base-En.tflite
WhisperEncoderInf QCS8450 (Proxy) QCS8450 Proxy QNN 383.819 ms 1 - 1441 MB FP16 NPU Use Export Script
WhisperEncoderInf Snapdragon X Elite CRD Snapdragon® X Elite QNN 174.12 ms 0 - 0 MB FP16 NPU Use Export Script
WhisperEncoderInf Snapdragon X Elite CRD Snapdragon® X Elite ONNX 202.131 ms 133 - 133 MB FP16 NPU Whisper-Base-En.onnx
WhisperDecoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 TFLITE 9.835 ms 5 - 31 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 QNN 5.8 ms 20 - 39 MB FP16 NPU Whisper-Base-En.so
WhisperDecoderInf Samsung Galaxy S23 Snapdragon® 8 Gen 2 ONNX 10.337 ms 11 - 446 MB FP16 NPU Whisper-Base-En.onnx
WhisperDecoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 TFLITE 7.583 ms 5 - 123 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 QNN 4.397 ms 186 - 258 MB FP16 NPU Whisper-Base-En.so
WhisperDecoderInf Samsung Galaxy S24 Snapdragon® 8 Gen 3 ONNX 8.155 ms 50 - 178 MB FP16 NPU Whisper-Base-En.onnx
WhisperDecoderInf Snapdragon 8 Elite QRD Snapdragon® 8 Elite TFLITE 7.295 ms 5 - 114 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf Snapdragon 8 Elite QRD Snapdragon® 8 Elite QNN 3.486 ms 18 - 84 MB FP16 NPU Use Export Script
WhisperDecoderInf Snapdragon 8 Elite QRD Snapdragon® 8 Elite ONNX 7.88 ms 50 - 155 MB FP16 NPU Whisper-Base-En.onnx
WhisperDecoderInf SA7255P ADP SA7255P TFLITE 36.508 ms 5 - 111 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf SA7255P ADP SA7255P QNN 26.501 ms 18 - 27 MB FP16 NPU Use Export Script
WhisperDecoderInf SA8255 (Proxy) SA8255P Proxy TFLITE 9.908 ms 5 - 31 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf SA8255 (Proxy) SA8255P Proxy QNN 4.359 ms 20 - 22 MB FP16 NPU Use Export Script
WhisperDecoderInf SA8295P ADP SA8295P TFLITE 12.313 ms 5 - 104 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf SA8295P ADP SA8295P QNN 5.765 ms 18 - 35 MB FP16 NPU Use Export Script
WhisperDecoderInf SA8650 (Proxy) SA8650P Proxy TFLITE 10.0 ms 5 - 28 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf SA8650 (Proxy) SA8650P Proxy QNN 4.213 ms 20 - 23 MB FP16 NPU Use Export Script
WhisperDecoderInf SA8775P ADP SA8775P TFLITE 12.335 ms 4 - 109 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf SA8775P ADP SA8775P QNN 5.42 ms 18 - 27 MB FP16 NPU Use Export Script
WhisperDecoderInf QCS8275 (Proxy) QCS8275 Proxy TFLITE 36.508 ms 5 - 111 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf QCS8275 (Proxy) QCS8275 Proxy QNN 26.501 ms 18 - 27 MB FP16 NPU Use Export Script
WhisperDecoderInf QCS8550 (Proxy) QCS8550 Proxy TFLITE 9.84 ms 6 - 31 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf QCS8550 (Proxy) QCS8550 Proxy QNN 4.237 ms 20 - 24 MB FP16 NPU Use Export Script
WhisperDecoderInf QCS9075 (Proxy) QCS9075 Proxy TFLITE 12.335 ms 4 - 109 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf QCS9075 (Proxy) QCS9075 Proxy QNN 5.42 ms 18 - 27 MB FP16 NPU Use Export Script
WhisperDecoderInf QCS8450 (Proxy) QCS8450 Proxy TFLITE 12.029 ms 5 - 116 MB FP16 NPU Whisper-Base-En.tflite
WhisperDecoderInf QCS8450 (Proxy) QCS8450 Proxy QNN 6.666 ms 20 - 91 MB FP16 NPU Use Export Script
WhisperDecoderInf Snapdragon X Elite CRD Snapdragon® X Elite QNN 3.797 ms 20 - 20 MB FP16 NPU Use Export Script
WhisperDecoderInf Snapdragon X Elite CRD Snapdragon® X Elite ONNX 9.212 ms 106 - 106 MB FP16 NPU Whisper-Base-En.onnx

Installation

Install the 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
------------------------------------------------------------
WhisperEncoderInf
Device                          : Samsung Galaxy S23 (13)   
Runtime                         : TFLITE                    
Estimated inference time (ms)   : 207.2                     
Estimated peak memory usage (MB): [0, 67]                   
Total # Ops                     : 419                       
Compute Unit(s)                 : GPU (408 ops) CPU (11 ops)

------------------------------------------------------------
WhisperDecoderInf
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 9.8                    
Estimated peak memory usage (MB): [5, 31]                
Total # Ops                     : 983                    
Compute Unit(s)                 : NPU (983 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
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S24")

# 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.

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.

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. 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

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