The autonomous oscillator model for speech synthesis is augmented by a nonlinear predic- tor to regenerate the modulated noiselike signal component of speech signals. The resulting `oscillator-plus-noise' model in combination with vocal tract modeling by linear prediction is able to regenerate the spectral content of stationary wide-band vowel signals with high fidelity. For adequate modeling of voiced fricatives the model is further extended by a second linear prediction path. With one and the same model not only sustained voiced and mixed excitation phonemes, but also unvoiced sounds can be regenerated faithfully.