Internet Low Bitrate Codec (iLBC) for VoIP

This example implements the Internet Low Bitrate Codec (iLBC) and illustrates its use. iLBC is designed for encoding and decoding speech for transmission via VoIP (Voice Over Internet Protocol).

VoIP

Voice over Internet Protocol is the family of technologies that allows IP networks to be used for voice applications such as telephony and teleconferencing. Compression is normally required to reduce the bandwidth requirements of these applications. For efficiency, VoIP is often implemented using the lightweight but unreliable User Datagram Protocol (UDP). Packet loss correction is needed to maintain received voice quality over lossy networks.

Basic iLBC Design and Performance

iLBC is designed for compression of speech to be transmitted over the Internet. Thus, its algorithms are only meant to cover the narrow frequency range of 90-4000 Hz and it implements perceptual coding tuned to normal speech. All input signals to the iLBC encoder must be Pulse Code Modulated (PCM) speech signals sampled at exactly 8000 Hz with 16-bit samples ranging from -32768 to +32767.

iLBC is defined for two different transmission rates, with a packet of data being encoded either after every 30ms or after every 20ms of speech. The advantage of encoding every 30ms is that the encoded data rate is lower: 13.33 kbit/sec as opposed to 15.20 kbit/sec for 20ms frames. However, encoding every 30ms leads to 50% more delay in the received speech, which can cause undesirable latency.

Since all inputs to iLBC must be 8000 Hz 16-bit PCM speech, the input rate is (8000 Hz) * (16 bits) = 128 kbit/sec. Thus, iLBC compresses the speech to 10.4% and 11.9% of the original data-rate for 13.33 kbit/sec and 15.20 kbit/sec modes, respectively.

In addition to encoding to low data transmission rates, iLBC provides a framework for easily implementing Packet Loss Correction (PLC) systems. The codec is meant for real-time speech over the Internet, but the Internet is subject to random delays in routing information in real-time, which renders many packets useless to the iLBC decoder. The job of a PLC is to interpolate the speech for missing packets based on the packets before and immediately after the missing one. Though iLBC does not define a specific PLC algorithm, this example implements a simple PLC for illustration.

The iLBC Example Model

The model shown below reads in a speech signal and, after passing through iLBC, plays the output with the default audio device.

Open and run the model.

Using the iLBC Example Model

The top level of this example model consists of just a handful of simple blocks. The basic operation is to load a speech signal and pass it to the iLBC Encoder block to convert it to a stream of iLBC packets. Next, the packets are sent through a simulated lossy channel, which causes random packets to be set to all zeros. Finally, the packets are sent to the iLBC Decoder block to be converted back into a speech signal, which is then played. In addition, there is a manual switch that can be toggled as the model runs to compare the original speech signal with the decoded signal.

Double clicking on the configuration block in the upper right corner of the model brings up a dialog, where it is possible to change the data transmission rate to one of the two iLBC modes (13.33 kbit/sec or 15.20 kbit/sec). The decoder's transmission rate must be set to the same as the encoder, or else an error will occur. In addition, the user may specify whether to use double or single precision for all internal calculations in the encoder and decoder.

Double clicking on the Lossy Channel subsystem brings up a dialog that allows the percentage of lost packets to be set. The iLBC Decoder's Packet Loss Concealment algorithm is tuned to correct for 0-10% packet loss. Packet loss rates higher than 10% will be easily audible.

The iLBC encoder and decoder blocks are implemented as subsystems in this model. In order to accommodate a level of reuse, they also make use of a example library, which can be found at dspilbclib. This library contains four helper blocks used by the encoder and decoder. Feel free to open the library and look under the blocks to see how iLBC was implemented in Simulink®.

Double clicking the Switch block will switch between the audio playback original speech signal and the iLBC decoded signal. The Time Scope block plots both the original speech and the iLBC decoded speech. Observe how closely the decoded speech matches the original speech.