In environments like open offices and co-working spaces, speech is often the main source of disruptive ambient noise, leading to reduced productivity, increased workload, and stress. Passive noise control techniques, such as office partitions and dividers, are often ineffective, making active noise control (ANC) headphones a more practical solution. However, the complexity and non-stationary nature of speech, combined with inherent constraints in ANC headphones, make it difficult to attenuate speech effectively.
 
These constraints are caused by delays in digital processing or acoustic propagation paths, which require the optimal filters to be non-causal. As a result, the anti-noise signal arrives too late at the eardrum to effectively cancel the ambient noise. The non-causality limits ANC performance and creates a prediction problem to compensate for the delay. Predicting non-stationary signals like speech presents significant challenges, leading to performance limitations.
 
This thesis addresses these challenges and aims to improve voiced speech attenuation by proposing new prediction methods and ANC system designs that leverage various characteristics of speech. These include designing optimal filters based on speech correlations, using sparse linear prediction modeling, applying harmonic-chirp model to account for non-stationarity of speech, and implementing multiple-frequency ANC with harmonic decomposition of speech.
 
The proposed methods outperform conventional ANC systems across a wide range of delays, achieving higher overall attenuation and extending voiced speech attenuation to higher frequencies. We conduct a perceptual study to assess the impact of these methods on user experience and explore whether objective metrics can predict subjective satisfaction with speech prediction-based ANC systems. The outcomes of this thesis can contribute to advancing ANC technology, enabling more effective and perceptually satisfying speech attenuation in real-world applications.

Reception
After the defence there will be a small reception at Fredrik Bajers Vej 7, B4-211