Skip to content

Latest commit

 

History

History
133 lines (93 loc) · 6.55 KB

README.md

File metadata and controls

133 lines (93 loc) · 6.55 KB
Raw

OnDevice Automatic Speech Recognition

Introduction

Automatic Speech Recognition (ASR) is one of the common and challenging Natural Language Processing tasks.

  • Current project is an sample Android application for OnDevice Automatic Speech Recognition using Qualcomm® Neural Processing SDK for AI framework.

  • We have used 1 Model in this Solution Wav2Vec2 Model

  • We Need to Give 1 Inputs: input_values of fixed Size(1,40000)

  • Models are small, efficient and mobile friendly Transformer model fine-tuned on librispeech dataset for ASR downstream task

  • In this project, we'll show how to efficiently convert, deploy and acclerate of these model on Snapdragon® platforms to perform Ondevice Automatic Speech Recognition(ASR).

Model Architecture

Prerequisites

  • Android Studio to import and build the project
  • Android NDK "r19c" or "r21e" to build native code in Android Studio
  • Python 3.8, PyTorch 1.10.1, Tensorflow 2.6.2, Transformers 4.18.0, Datasets 2.4.0 to prepare and validate the model
    (above mentioned Python packages version and Android Studio version is just a recommendation and is not a hard requirement. Please install SDK dependencies in Python 3.8 virtual environment)
  • Qualcomm® Neural Processing Engine for AI SDK v2.x.x and its dependencies to integrate and accelerate the network on Snapdragon
    (During developement of this tutorial, the AI SDK recommends Python 3.8 version and is subject to change with future SDK releases. Please refer SDK Release Notes.)

Quick Start

1. Model Preparation

Please go to wav2vec2_model_generation/accuracy_analyzer.ipynb file to generate the models and put the models inside app/src/main/assets folder,Before running this Notebook Setup your SNPE Environment from the below code.

1.2 Setup the Qualcomm® Neural Processing SDK Environment:

source <snpe-sdk-location>/bin/envsetup.sh

1.3 Convert generated frozen graph into DLC (Deep Learning Container):

snpe-onnx-to-dlc -i model.onnx -d input_values 1,40000 -o wav2vec2_fp32.dlc
(you can add outputTensor name in the above command, please check SNPE Document for more information. Please check once by visualizing graph using Netron viewer or any other visualization tools )

1.4 Quantization with caching of DLC (for optimizing model loading time on DSP accelerator)

snpe-dlc-quantize --input_dlc models/wav2vec2_fp32.dlc --input_list list.txt  --optimizations cle --output_dlc models/wav2vec2_w8a16.dlc --enable_htp --htp_socs sm8650 --weights_bitwidth 8 --act_bitwidth 16

  • Here In this Android App We've used wav2vec2_fp32.dlc
  • You can change for desired DLC from app/src/main/cpp/native_lib.cpp folder

4. Build and run with Android Studio

Add AI SDK libs and generated DLC into app assets, jniLibs and cmakeLibs directory:

  • Take SNPE_ROOT/lib/android/snpe-release.aar file, unzip it
  • Then create app/src/main/jniLibs folder and paste everything from jni/arm64-v8a which is extracted from snpe-release.aar to this app/src/main/jniLibs/arm64-v8a folder.
  • Take SNPE_ROOT/lib/hexagon-v75 if you want to run it on Snapdragon 8th gen 3 device
  • Take SNPE_ROOT/lib/hexagon-v73 if you want to run it on Snapdragon 8th gen 2 device.
  • If build process fails with libSNPE.so duplication error, then please change its path from "jniLibs" to "cmakeLibs" as follows : ${CMAKE_CURRENT_SOURCE_DIR}/../cmakeLibs/arm64-v8a/libSNPE.so in app/src/main/cpp/CMakeList.txt under target_link_libraries section and delete libSnpe.so from "jniLibs" directory.

Debug Tips

  • After installing the application, if it is crashing, try to collect the logs from QIDK device.

  • To collect logs run the below commands.

    • adb logcat -c
    • adb logcat > log.txt
    • Now, run the app. Once, the app has crashed do Ctrl+C to terminate log collection.
    • log.txt will be generated in current folder.
    • Search for the keyword "crash" to analyze the error.

  • On opening the app, if Unsigned or Signed DSP runtime is not getting detected, then please search the logcat logs with keywork dsp for the FastRPC errors.

  • DSP runtime may not get detected due to SE Linux security policy in some Android builds. Please try out following commands to set permissive SE Linux policy.

adb disable-verity
adb reboot
adb root
adb remount
adb shell setenforce 0
// launch the application

Result

Qualcomm® Neural Processing SDK C++ APIs JNI Integration

Please refer to SDK Native application tutorial : https://developer.qualcomm.com/sites/default/files/docs/snpe/cplus_plus_tutorial.html

Credits

The pre-trained model is from HuggingFace Repository by MRM8488 https://huggingface.co/docs/transformers/model_doc/wav2vec2

References

Qualcomm Neural Processing SDK is a product of Qualcomm Technologies, Inc. and/or its subsidiaries.