Noise reduction in speech signals using adaptive filters

The need for communication in humanity and the current era has made the definition, transmission, and improvement of speech signals a crucial topic. Reducing, separating, and resolving noise in speech signals is essential, especially in noisy, crowded environments. This ensures that vital information and data can be recorded and extracted based on the intended recipient's voice. The field of speech signal processing encompasses modeling the voice, identifying its various characteristics and parameters, recognizing speech or speakers, and creating artificial speech. It also involves developing methods for compressing and decompressing sound, encoding voice, and improving noisy sounds. Noise can generally be defined as any kind of sound that disturbs people. In this context, various types of noise exist, including acoustic, environmental, electrical, and technological noise, as well as electromagnetic, communication line, quantum, and noise in sound communication. Noise in audio communications can distort or render the intended sound unintelligible, making communication difficult or ineffective. Noise plays a significant role in industrial, scientific, and technological applications. Therefore, controlling and reducing noise is important to enhance efficiency in communication and other fields. When considering noise in terms of signals, it can sometimes be confined to a specific frequency range, while other times its position in the frequency spectrum may change over time. To reduce fixed-frequency noise, filters that block specific frequency bands (band-stop filters) are typically used. However, if the frequency of the noise changes over time, adaptive filters are preferred. Adaptive filters continuously update filter coefficients based on the characteristics of the noise. The use of adaptive filters is considered an effective method for reducing noise in speech signals. These filters can adapt to the changing characteristics of the noise and effectively suppress unwanted components. In this study, the topic of noise reduction in speech signals is addressed by first examining sound and sound signals, followed by the formation of speech and its signals. Subsequently, noise and noise in speech signals are discussed. In the next section, the structure of adaptive filters, their usage advantages, and their application types for noise in speech signals are examined. A method is followed where appropriate algorithms are developed to reduce noise in speech signals using adaptive filters, and their tests are conducted on a Python program. Speech recordings made in two different environments are processed through various filters, and their noise reduction successes are analyzed. The results show that at a low SNR level of 0 dB, both filters exhibit limited performance. With an increase in the SNR level to 5 dB, improvements in the performance of both filters are observed. At an even higher SNR level of 10 dB, significant improvements in the performance of both filters are noted. In a traffic noise environment, the performance of the Kalman filter is generally higher compared to the shopping mall environment. However, at low SNR levels, the RLS filter performs better than the Kalman filter, especially in the shopping mall environment.