Speech Reinforcement and Acoustic Feedback Cancellation

Speech reinforcement (SR) techniques aim to increase the speech intelligibility in adverse environment where the communication is problematic due to the presence of noise or considerable attenuation of speech signal. Thus, an SR system is composed by one microphone, an amplifier and a loudspeaker at least. In addition there are the algorithms for process the signal between the acquisition and the reproduction stage.

Different scenarios can be found depending on the size of the room and the amount of reverberation but all the different situations are characterized by the presence of acoustic feedback. A typical scenario with acoustic feedback is a signal received by the microphone that is amplified and played back by the loudspeaker. Here, the sound from the loudspeaker can then be received by the microphone again, amplified further, and then passed out through the loudspeaker again, creating instability.

Different approaches attempting to solve the acoustic feedback problem have been proposed in literature.  One of the earliest approach is the phase-modulating feedback control (PFC), almost unused due to its poor performance. Belonging to the class of gain reduction methods, notch-filter based howling suppression (NHS) represents a traditional solution, well-known for its robustness. It takes the system back to stability using notch filters, aiming to reduce the forward path gain in narrow frequency bands around those frequencies with a loop gain close to unity. This solution is usually unable to prevent instability, which is tackled only after a detection phase of about 0.5 seconds, and therefore is said to be reactive. On the other hand, adaptive feedback cancellation (AFC) is an emerging technique belonging to the class of room modeling methods, based on the estimation of the acoustic feedback path. This allows the system to get an estimate of the feedback signal component which is then subtracted from the microphone signal. Depending on the quality of the estimation, a nearly complete elimination of the acoustic coupling can be achieved. In this way, AFC using a prediction-error-method (PEM) approach is a promising proactive solution. In particular the PEM-AFROW algorithm revealed to have relevant feedback cancellation capabilities and a certain flexibility to integration with other algorithms to face more general case studies, i.e. the presence of background noise  and of time-varying impulse responses.

Nevertheless, in some cases, the non-perfect tracking behavior of the PEM-AFROW algorithm may lead to an unstable system, especially when operating close to the stability treshold. Large changes in the acoustic feedback path, for example, have shown to be quite troublesome. The Suppressor-PEM approach copes with this problem, employing a feedback suppression filter in cascade with the PEM-AFROW algorithm, leading to a more robust solution.

A source code  in C of a single channel implementation of this algorithm is available below. It has been tested in a real environment on an OMAP-L138 experimenter kit (by Texas Instruments) and integrated with a NU-Tech based test bench where an automatic gain controller (AGC) has been implemented.

A3LAB research work is oriented to study and develop robust AFC algorithms for real-world applications. In particular, the  intra-cabin communication task (specially oriented to the automotive field) has been addressed and valid solutions already proposed in the scientific literature.