Pre-emphasis is a very simple signal processing method which increases the amplitude of high frequency bands and decrease the amplitudes of lower bands. In simple form it can be implemented as

[math]y_t = x_t - \alpha x_{t-1}[/math]

Pre-emphasis is not really required, it was introduced many years ago when limited computing resources forced developers to create tricky methods. It was noted that higher frequencies are more important for signal disambiguation than lower frequencies. In those days it would be easier to apply analog preemphasis filter to get slightly better results so the pre-emphasis become popular. Another good property of preemphasis is that it helps to deal with DC offset which is often present in recordings and thus it can improve energy-based voice activity detection.

Modern speech recognition do not require pre-emphasis. Preemphasis is compensated on later stages with channel normalization like cepstral mean normalization so it should have no effect at all. It is an artifact from a older system design.