Context classification from audio data using machine learning

Société : MOVEA (Groupe TDK)

Company Overview:
American R&D Center (ARDC) is a central R&D organization of the TDK Group. Its mission is to contribute to the design of new products within TDK with the ambition to create new product lines. ARDC is working closely with InvenSense Inc., a TDK Group Company, who is the world’s leading provider of MEMS sensor platforms. InvenSense’s vision of Sensing Everything™ targets the consumer electronics and industrial markets with integrated Motion, Sound, UltraSound, Pressure, Gas solutions.
InvenSense’s motion tracking, audio, ultrasound, and services can be found in many of the world’s largest and most iconic brands including smartphones, tablets, wearables, drones, gaming devices, internet of things, automotive products, and remote controls for smart TVs.

Both ARDC & InvenSense are headquartered in San Jose, CA, and have offices in Boston, China, Taiwan, Korea, Japan, France, Slovakia, and Italy.

Description du poste

Internship Proposal :
Context classification from audio data using machine learning Detailed description of the internship:
Smart applications and smart devices are at the brink to become part of our everyday life. In order to improve user experience, the smart applications as of today do not exploit the environmental context in a satisfactory manner. Examples are:
• indoor / outdoor context detection for enabling / disabling wind suppression for ear buds.
• Empty room / crowded room / room with few occupants for adapting AC and lighting in a smart building.

The goal of this internship is to explore such context detection using audio data.
The first step would be to do a literature review of existing solutions, focusing on neural networks with recurrent layers such as LSTM and GRU, temporal convolution layers such as TCN and attention layers. The internship requires in addition to this theoretical part a practical part where a context classifier is built.

The student will become familiar with our acquisition software, database collection protocol, audio pre-processing, neural network training and testing, evaluation metrics, reporting of results including memory consumption, response latency, accuracy, false alarm rate.
The team has a focus on embedded solutions. If possible, the student should take into account limited memory and power constraints.

Tasks:
• Learning: get familiar with in house audio hardware.
• Learning: get familiar machine learning methods.
• Characterization: database collection and annotation.
• Implementation: prototype solution using Python, matlab or C.
• Evaluation: classification accuracy, false alarm rate, memory consumption, response latency. Describe limitations of the method.
• Demo: context classifier using audio data.

Profil recherché

Contract: Professional Internship
Qualification required: Bac+5 program
Dates: To be defined
Duration: 6 months
Where: Grenoble

Skills required :
Programming in C/Matlab/Python. Good knowledge of signal processing and estimation theory. Pugnacity and creativity are paramount skills for this challenging problem.

Voir le fichier joint