Deep Neural Networks with Acoustics and Pitch related Features

Client Overview

US based leading semiconductor manufacturer offering microcontrollers and processors for sensors, analog ICs and connectivity.

Business Need

  • The client was looking for a technology partner to help them in proving the application of newly developed chipsets for machine learning at edge to be launched in the market

VOLANSYS Contribution

  • Designed the voice-based gender detection application using supervised machine learning algorithm – Depth-wise separable convolution neural network
  • DS-CNN is based on MobileNet architecture that is best suited for memory constrained devices
  • Extracted audio features using Mel Frequency Cepstral Coefficients technique
  • Developed model using TensorFlow framework and trained model with extracted features from 9000 audio files
  • Developed application on NXP i.MXRT600, performed data pre-processing and audio feature extraction (MFCC) of the real-time audio samples on Hi-Fi 4 DSP and the inference of the Deep Learning model was deployed on ARM Cortex-M33
  • Performance evaluation using real human voice input through microphone

Solution Diagram


Benefits Delivered

  • Accelerated client’s product launch timeline by 20% with years of expertise in machine learning domain
Similar Success Stories