This paper presents a Hybrid RLS-NLMS adaptive algorithm for remote cancellation of multi-tone noise signals using a directional ultrasonic loudspeaker. It combines the stable and computationally efficient but slow converging traditional Normalized Least Mean Squares (NLMS) algorithm with a fast converging yet computationally complex Recursive Least Squares (RLS) algorithm. This gives a fast converging, stable and less computationally complex Hybrid RLS-NLMS algorithm, which gives lesser residual error than compared to a NLMS algorithm and lower overall computational complexity than a RLS algorithm individually. The anti-noise signal generated by proposed algorithm modulates the 60 kHz carrier frequency of the directional ultrasonic loudspeaker which serves as the noise canceling loudspeaker. Automatic demodulation of the ultrasonic output signal cancels the noise signal at a distant target point within the loudspeaker output pattern. Experimental results for real time implementation of the Hybrid algorithm to cancel various multi-tone noise signals using the ultrasonic loudspeaker are presented.