edVirtus Reference

What Is Regular Pulse Excitation?

What Is RPE?

Regular Pulse Excitation (RPE) is a low-bit-rate speech-coding technique that represents the excitation signal of the human vocal tract using a series of regularly spaced pulses. Developed as an improvement to Linear Predictive Coding (LPC), RPE produces more natural-sounding speech than earlier vocoders while maintaining relatively low transmission bit rates. It became widely known through its adoption in the original Global System for Mobile Communications (GSM) full-rate speech codec.

Like other predictive speech coders, RPE is based on the source-filter model of speech production. The vocal tract is represented by a mathematical filter derived using Linear Predictive Coding (LPC), while the excitation signal models the airflow generated by the vocal cords. Instead of transmitting the speech waveform directly, only the filter parameters and a simplified description of the excitation are transmitted.

The distinguishing feature of RPE is that the excitation is represented by pulses occurring at regular intervals rather than at arbitrary positions. The encoder selects the pulse sequence that best approximates the original speech while requiring relatively few bits to describe it. Restricting the pulses to fixed positions greatly simplifies both the encoding process and the amount of information that must be transmitted.

A useful analogy is creating a sketch using graph paper. Rather than placing dots anywhere on the page, the artist places them only at the intersections of the grid. Although this slightly limits flexibility, it makes the drawing process much simpler while still capturing the essential features of the image. RPE applies the same principle to the excitation signal by restricting pulses to regularly spaced positions.

In practical implementations, RPE is often combined with Long-Term Prediction (LTP), producing the well-known RPE-LTP algorithm. Long-term prediction exploits the periodic nature of voiced speech by modelling pitch, while RPE represents the remaining excitation. Together, these techniques provide significantly better speech quality than conventional LPC while maintaining bit rates suitable for mobile communication systems.

The best-known application of RPE was the original GSM Full-Rate speech codec, which operated at approximately 13 kbps. During the early years of digital cellular telephony, RPE-LTP provided an excellent balance between speech quality, computational complexity, and transmission bandwidth, enabling efficient voice communication over the limited radio spectrum available to GSM networks.

It is important to distinguish Regular Pulse Excitation (RPE) from Multipulse Excitation (MPE). In MPE, the excitation pulses may occur at arbitrary positions within the speech frame, providing greater modelling flexibility but requiring more complex encoding. In RPE, the pulses are constrained to regular intervals, simplifying implementation and reducing the number of transmitted bits. Later techniques such as Code-Excited Linear Prediction (CELP) further improved speech quality by selecting excitation sequences from a predefined codebook rather than explicitly specifying pulse positions.

Today, RPE is primarily of historical importance because it formed the basis of one of the world's first widely deployed digital mobile telephone speech codecs. Although newer techniques such as CELP and its derivatives now provide superior speech quality and compression efficiency, RPE demonstrated that high-quality digital speech could be achieved at relatively low bit rates, helping to make large-scale digital mobile communications commercially practical.