AN2113 Freescale Semiconductor / Motorola, AN2113 Datasheet - Page 5

• Manufacturer Part Number: AN2113
• Description: AN2113 Multichannel Voice Coding System on the RTXC Operating System
• Manufacturer: Freescale Semiconductor / Motorola

Input Speech
both a codebook index value indicating which excitation parameter is appropriate and a codebook gain
indicating the strength of the excitation. The tables for IS-96-A require 6622 words of data space when
implemented on the DSP56300 family.
IS-96-A is a variable-rate coder, so it senses when speech activity lessens and transmits less information
during that time. This lowers the bandwidth requirements during inactive speech times and improves the
efficiency of the system in general. Therefore, theoretically a given base station can handle more channels
based on the statistical knowledge that not every user is at maximum speech data capacity at a given time.
Also, depending on the transmission method, the variable rate cuts down the noise for other users. The
maximum rate for IS-96-A is 8 kbps and the minimum rates are 4 kbps (1/2 rate), 2 kbps (1/4 rate), and 0.8
kbps (1/8 rate). The encoder sends information to the decoder when it adjusts the rate. It jumps only one
rate per given sample time, so the encoder does not send a full transmission rate sample and then sense
complete silence and drop down to the minimum transmission rate on the next sample. Instead, it cycles
through each rate as appropriate for a given sample time and does not adjust by more than one rate at a
time. This behavior helps to maintain the quality of the speech.
The encoder analyzes the input speech and transmits a set group of speech parameters to the decoder.
These parameters include coefficients related to the formants that determine the resonant frequencies of the
vocal tract at a given time. The encoder also transmits the codebook gain and index, pitch information, and
parity check bits. The encoder determines these parameters during the process shown in Figure 4. In
IS-96-A, the encoder implements a search procedure to recreate the input speech by comparing it to the
output of the synthesizer in the encoder. For each received input sequence, the encoder attempts to
synthesize the speech, comparing its output with the input speech and calculating a weighted error value.
Once this error is sufficiently minimized, the parameters that create the “best” synthesized speech are
transmitted to the decoder.
The decoder receives the parameters transmitted by the encoder and reproduces the speech so that the
person listening on the receive end can understand the person speaking. The decoding process does not
require the analysis of the speech that the encoder must complete, so decoding requires less processing
power for a system than encoding. Figure 5 shows the steps required for decoding in IS-96-A.The first
filter, called the long-term filter, reconstructs the long-term pitch periodicities of the speech in the
excitation signal. The second, called the short-term filter, models the spectral shape of the speech.
3. http://www.msdmag.com/frameindex.htm?98/9806art4.htm.
Codebook
Code
Gain
Freescale Semiconductor, Inc.
x
For More Information On This Product,
Figure 4. IS-96 Encode Block Diagram
Long-Term
Multichannel Voice Coding System
Go to: www.freescale.com
B(z)
Short-Term
A(Z)
s1(n)
-
s(n)
W(z)
Weighted Error
Calculator
Energy
Voice Coding
3
5