We investigate characteristics and benefits of speech advisories with the goal of enhancing human-machine interaction through their use. Spontaneous speech data was collected using a simple task. Characteristics of advisory speech are (1) utterances usually include redundant information, (2) when redundant information is included, main information is emphasized by changing F0 value, (3) when communication goes well, utterances often include only necessary information. To confirm the benefits of speech advisories, we designed an experiment in which a computer monitors a human operator and issues verbal advice when the operator makes a mistake. Results are (1) operators were found to divide into two groups; i.e., preferred silence and preferred audio output, (2)they preferred synthetic speech to human speech; human speech made them feel as if they were being personally monitored.