ML-KWS-for-FPGA

嵌入式计算系统及SoC软硬件协同设计课程大作业。

Usage

1. Download Google speech command dataset
2. Put the dataset outside this project, specifically, at ../speech/
3. git clone https://github.com/zYeoman/ML-KWS-for-FPGA
4. CD in ML-KWS-for-FPGA and make and run kws

wget http://download.tensorflow.org/data/speech_commands_v0.01.tar.gz
tar xzvf speech_commands_v0.0.1.tar.gz speech
git clone https://github.com/zYeoman/ML-KWS-for-FPGA && cd ML-KWS-for-FPGA
make
# Run CNN model for the first 10 test.
./kws cnn 10
# Run CRNN model for all test
./kws crnn
# Run CNN_Q(Quantization CNN)
./kws cnn_q

Hardware

In Vivado HLS, use kws(uint32_t*, int32_t*) or kws_q or kws_crnn as top function, test.cpp as test source. In Vivado, Open xillinux blockdesign, and connect the IP core and Xillybus as shown belown.

Files

./bit           Bitstream
./Makefile      Makefile
./silence.wav   Testfile
./include       Header And Model Parameters
./src           Source