Whisper-Small-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-Small-En found here.

This repository provides scripts to run Whisper-Small-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: small.en
- Input resolution: 80x3000 (30 seconds audio)
- Mean decoded sequence length: 112 tokens
- Number of parameters (WhisperEncoder): 102M
- Model size (WhisperEncoder): 390 MB
- Number of parameters (WhisperDecoder): 139M
- Model size (WhisperDecoder): 531 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	55.414 ms	14 - 41 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	45.791 ms	16 - 417 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	40.412 ms	21 - 277 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	SA7255P ADP	SA7255P	TFLITE	120.347 ms	16 - 267 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	SA8255 (Proxy)	SA8255P Proxy	TFLITE	55.503 ms	16 - 42 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	SA8295P ADP	SA8295P	TFLITE	51.275 ms	16 - 248 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	SA8650 (Proxy)	SA8650P Proxy	TFLITE	55.643 ms	16 - 41 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	SA8775P ADP	SA8775P	TFLITE	55.559 ms	16 - 267 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	QCS8275 (Proxy)	QCS8275 Proxy	TFLITE	120.347 ms	16 - 267 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	55.652 ms	16 - 41 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	QCS9075 (Proxy)	QCS9075 Proxy	TFLITE	55.559 ms	16 - 267 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperDecoder	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	58.245 ms	14 - 402 MB	FP16	NPU	Whisper-Small-En.tflite
WhisperEncoder	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	696.187 ms	110 - 134 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	934.985 ms	106 - 200 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	449.288 ms	109 - 140 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	SA7255P ADP	SA7255P	TFLITE	4466.415 ms	101 - 133 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	SA8255 (Proxy)	SA8255P Proxy	TFLITE	701.364 ms	97 - 118 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	SA8295P ADP	SA8295P	TFLITE	655.091 ms	109 - 140 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	SA8650 (Proxy)	SA8650P Proxy	TFLITE	683.542 ms	102 - 208 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	SA8775P ADP	SA8775P	TFLITE	1290.925 ms	108 - 139 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	QCS8275 (Proxy)	QCS8275 Proxy	TFLITE	4466.415 ms	101 - 133 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	954.302 ms	56 - 111 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	QCS9075 (Proxy)	QCS9075 Proxy	TFLITE	1290.925 ms	108 - 139 MB	FP16	GPU	Whisper-Small-En.tflite
WhisperEncoder	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	1036.055 ms	83 - 180 MB	FP16	GPU	Whisper-Small-En.tflite

Installation

Install the package via pip:

pip install "qai-hub-models[whisper-small-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_small_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_small_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_small_en.export

Profiling Results
------------------------------------------------------------
WhisperDecoder
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 55.4                   
Estimated peak memory usage (MB): [14, 41]               
Total # Ops                     : 2573                   
Compute Unit(s)                 : NPU (2573 ops)         

------------------------------------------------------------
WhisperEncoder
Device                          : Samsung Galaxy S23 (13)   
Runtime                         : TFLITE                    
Estimated inference time (ms)   : 696.2                     
Estimated peak memory usage (MB): [110, 134]                
Total # Ops                     : 911                       
Compute Unit(s)                 : GPU (900 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_small_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-Small-En's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of Whisper-Small-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.