Student Theses - Details
Characterization of Earbud-Mounted Accelerometers as Sensors for Speech Enhancement
Supervisors: Christoph Weyer
Area of work: Speech Enhancement, Headphone Technology, Acoustics
Tools: Matlab / Python, Acoustic Measurements, Hardware, Statistical Signal Analysis
Category: Bachelor Thesis / Master Thesis
Some recent bluetooth earbuds are equipped with high-bandwith accelerometers, which can receive speech of the user as vibrations transmitted through the head. At the same time, these accelerometers are very robust against outside acoustical and wind noise. Thus, they have great potential for speech enhancement of the user's voice.
While research exists regarding speech enhancement based on body conducted signals, most of it does not consider earbud-mounted accelerometers. Thus, many basic questions remain before employing such a sensor. The main goal of this thesis is to investigate the basic characteristics of an accelerometer setup as well as the produced signals and how they can be used in speech enhancement.
First, relevant literature shall be reviewed to gather an understanding of the relevant acoustic, as well as existing research and patents on earbud-mounted accelerometers. Then measurements shall be carried out to characterize the speech signal received by the accelerometer. The measurements shall be conducted using commercial headphones and different minituarized accelerometers. Different mounting positions, fits, excitation signals, outside noise, and possibly different users shall be measured. Then, the measured signals shall be compared by a statistical analysis to quantify the amount of information gained by the accelerometers, using measures such as group delay, cross power spectral density and coherence, or mutual information. This may include the joined analysis of multiple accelerometers or axis.