Hugging Face's logo

qualcomm
/
Whisper-Small-V2

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

Whisper-Small-V2: Optimized for Mobile Deployment

Transformer-based automatic speech recognition (ASR) model for multilingual transcription and translation available on HuggingFace

HuggingFace Whisper-Small ASR (Automatic Speech Recognition) model is a state-of-the-art system designed for transcribing spoken language into written text. 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 model is an implementation of Whisper-Small-V2 found here.

This repository provides scripts to run Whisper-Small-V2 on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

  • Model Type: Model_use_case.speech_recognition
  • Model Stats:
    • Model checkpoint: openai/whisper-small
    • Input resolution: 80x3000 (30 seconds audio)
    • Max decoded sequence length: 200 tokens
    • Number of parameters (HfWhisperEncoder): 102M
    • Model size (HfWhisperEncoder): 391 MB
    • Number of parameters (HfWhisperDecoder): 139M
    • Model size (HfWhisperDecoder): 533 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
HfWhisperEncoder float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 3341.744 ms 109 - 159 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 427.624 ms 0 - 10 MB NPU Use Export Script
HfWhisperEncoder float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 1005.626 ms 83 - 280 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 312.338 ms 0 - 356 MB NPU Use Export Script
HfWhisperEncoder float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 757.256 ms 14 - 220 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 136.071 ms 4 - 6 MB NPU Use Export Script
HfWhisperEncoder float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 1417.847 ms 100 - 149 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 157.0 ms 1 - 10 MB NPU Use Export Script
HfWhisperEncoder float SA7255P ADP Qualcomm® SA7255P TFLITE 3341.744 ms 109 - 159 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float SA7255P ADP Qualcomm® SA7255P QNN 427.624 ms 0 - 10 MB NPU Use Export Script
HfWhisperEncoder float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 771.963 ms 18 - 158 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 138.118 ms 1 - 9 MB NPU Use Export Script
HfWhisperEncoder float SA8295P ADP Qualcomm® SA8295P TFLITE 727.676 ms 109 - 158 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float SA8295P ADP Qualcomm® SA8295P QNN 234.176 ms 1 - 17 MB NPU Use Export Script
HfWhisperEncoder float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 753.393 ms 40 - 156 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 137.6 ms 1 - 3 MB NPU Use Export Script
HfWhisperEncoder float SA8775P ADP Qualcomm® SA8775P TFLITE 1417.847 ms 100 - 149 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float SA8775P ADP Qualcomm® SA8775P QNN 157.0 ms 1 - 10 MB NPU Use Export Script
HfWhisperEncoder float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 746.335 ms 6 - 270 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 137.066 ms 0 - 27 MB NPU Use Export Script
HfWhisperEncoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 557.898 ms 110 - 300 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 108.082 ms 0 - 467 MB NPU Use Export Script
HfWhisperEncoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 287.423 ms 227 - 2534 MB NPU Whisper-Small-V2.onnx
HfWhisperEncoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 621.432 ms 110 - 155 MB GPU Whisper-Small-V2.tflite
HfWhisperEncoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 82.303 ms 0 - 440 MB NPU Use Export Script
HfWhisperEncoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 208.333 ms 230 - 2106 MB NPU Whisper-Small-V2.onnx
HfWhisperEncoder float Snapdragon X Elite CRD Snapdragon® X Elite QNN 132.231 ms 0 - 0 MB NPU Use Export Script
HfWhisperEncoder float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 369.456 ms 264 - 264 MB NPU Whisper-Small-V2.onnx
HfWhisperDecoder float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 31.514 ms 14 - 565 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN 17.731 ms 48 - 59 MB NPU Use Export Script
HfWhisperDecoder float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 22.866 ms 14 - 716 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN 17.944 ms 54 - 127 MB NPU Use Export Script
HfWhisperDecoder float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 21.28 ms 10 - 37 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN 11.601 ms 54 - 57 MB NPU Use Export Script
HfWhisperDecoder float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 23.127 ms 14 - 565 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN 13.132 ms 55 - 64 MB NPU Use Export Script
HfWhisperDecoder float SA7255P ADP Qualcomm® SA7255P TFLITE 31.514 ms 14 - 565 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float SA7255P ADP Qualcomm® SA7255P QNN 17.731 ms 48 - 59 MB NPU Use Export Script
HfWhisperDecoder float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 21.07 ms 14 - 41 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN 12.089 ms 60 - 63 MB NPU Use Export Script
HfWhisperDecoder float SA8295P ADP Qualcomm® SA8295P TFLITE 22.268 ms 10 - 501 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float SA8295P ADP Qualcomm® SA8295P QNN 14.142 ms 55 - 70 MB NPU Use Export Script
HfWhisperDecoder float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 21.474 ms 14 - 43 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN 12.008 ms 60 - 63 MB NPU Use Export Script
HfWhisperDecoder float SA8775P ADP Qualcomm® SA8775P TFLITE 23.127 ms 14 - 565 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float SA8775P ADP Qualcomm® SA8775P QNN 13.132 ms 55 - 64 MB NPU Use Export Script
HfWhisperDecoder float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 21.154 ms 14 - 42 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN 11.897 ms 52 - 75 MB NPU Use Export Script
HfWhisperDecoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 16.844 ms 3 - 766 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN 9.519 ms 60 - 181 MB NPU Use Export Script
HfWhisperDecoder float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 20.449 ms 0 - 774 MB NPU Whisper-Small-V2.onnx
HfWhisperDecoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 14.432 ms 6 - 546 MB NPU Whisper-Small-V2.tflite
HfWhisperDecoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN 8.214 ms 59 - 153 MB NPU Use Export Script
HfWhisperDecoder float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 16.787 ms 147 - 401 MB NPU Whisper-Small-V2.onnx
HfWhisperDecoder float Snapdragon X Elite CRD Snapdragon® X Elite QNN 10.057 ms 60 - 60 MB NPU Use Export Script
HfWhisperDecoder float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 19.893 ms 227 - 227 MB NPU Whisper-Small-V2.onnx

Installation

Install the package via pip:

pip install "qai-hub-models[whisper-small-v2]"

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_v2.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_v2.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_v2.export

Profiling Results
------------------------------------------------------------
HfWhisperEncoder
Device                          : cs_8275 (ANDROID 14)                  
Runtime                         : TFLITE                                
Estimated inference time (ms)   : 3341.7                                
Estimated peak memory usage (MB): [109, 159]                            
Total # Ops                     : 1806                                  
Compute Unit(s)                 : npu (0 ops) gpu (1798 ops) cpu (8 ops)

------------------------------------------------------------
HfWhisperDecoder
Device                          : cs_8275 (ANDROID 14)                  
Runtime                         : TFLITE                                
Estimated inference time (ms)   : 31.5                                  
Estimated peak memory usage (MB): [14, 565]                             
Total # Ops                     : 3136                                  
Compute Unit(s)                 : npu (3136 ops) gpu (0 ops) cpu (0 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_v2 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-Small-V2's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of Whisper-Small-V2 can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
161
Inference Providers NEW
Automatic Speech Recognition
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support