When a speaker hears his own voice fed back to him with altered intensity or timing, he adjusts the properties of his speech to compensate for the apparent distortion of his message. These adjustments appear to be designed to provide the listener with intelligible messages rather than to provide the speaker with some acoustic feedback required for elaborating speech. When it comes to a third type of sidetone distortion, changes in spectral properties or timbre, much is claimed but little is known. On the one hand, the speaker might modify the source spectrum or the vocal tract transfer function so as to compensate for changes in the apparent timbre of his voice, pre-emphasizing high frequencies under low-pass sidetone filtering, and conversely. On the other hand, reductions in regions of the sidetone spectrum have been claimed to yield corresponding reductions in the voice spectrum. In fact the present experiment found no appreciable changes in the long-term spectra of sentences read by four speakers under seven conditions of sidetone filtering, once fluctuations in sidetone level were controlled. The speakers proved to be more intelligible, however, when they heard their voices filtered while speaking than when they did not.