This work proposes a computational algorithm which monitors voice/audio signals demodulated from a FM receptor and detects whether they are intelligible or not. Data analytics applications which require the continuous storage of radio broadcasted audio signals into a database can benefit from this algorithm. In many instances, the broadcasted signals arrive at the receptor with heavy distortion and noise content, limiting the data analysis due to poor data quality. Moreover, radio spectrum supervisory agencies can also take advantage of this work, since broadcasted signals can be efficiently and continuously monitored to detect whether a broadcaster has stopped transmitting for an extended period. First, the algorithm processes the demodulated signals block by block, extracting its MFCC coefficients, spectral centroid, the arithmetic and geometric means of the frequency magnitude spectrum and the zero-crossing rate in the time domain. Then, these parameters enter a classification algorithm based on three successive support vector machines (SVM), which output one of four possible classes for each block: intelligible clean signal, intelligible noisy signal, unintelligible noisy signal, and noise/silence signal. The algorithm has a 99.85% accuracy for intelligible clean signal versus unintelligible noisy/noise/silence signals; 97.34% accuracy for intelligible noisy signal versus noise/silence signals; and 96.36% accuracy for intelligible voice versus noise/silence.