Abstract: In a speech coder, a sampled analog signal is filtered by a short-term prediction filter. The result, a segmented residual signal, is transformed from a time domain to a frequency domain into several frequency components, each having a frequency-component amplitude. If a number of new amplitudes is calculated by combining the several frequency-component amplitudes, such that the number of new amplitudes is smaller than the several frequency-component amplitudes, a more efficient coder is created. The reduction of the quality of speech coding, due to loss of information, could be decreased if this calculation is based on the so-called Bark scale (critical frequency bands). In a corresponding speech decoder, at the hand of the number of new amplitudes several new frequency-component amplitude are calculated (the number of new amplitudes being smaller than the several new frequency-component amplitudes), which then are inverse transformed from a frequency domain to a time domain into new subsegments.

Abstract: A method and an arrangement are described by means of which a pitch in a signal (s(t)) can be determined. In fact the probability density function (pdf) of the pitch is determined as a function of frequency. A pitch can then be derived from this probability density function. First a frequency spectrum A(f) is derived from a signal portion situated in a specific time interval T. Then, the probability contribution (K.sub.1i, K.sub.1j, . . . . ) is determined for a number of frequency components (f.sub.i, f.sub.j, . . . ), stating the relative probability that a pitch at a frequency (f.sub.1) is the result of a frequency component (f.sub.i, f.sub.j, . . . ). The probability contributions (K.sub.1i, K.sub.1j, . . . ) are multiplied by one another to obtain a probability density (K.sub.f1). This procedure is iterated for successive frequencies (f.sub.2, . . . ) to obtain at least a second probability density. The probability densities K.sub.f1, K.sub.f2, . . .