ISD-T360SB Datasheet, PDF (50/109 Pages) List of Unclassifed Manufacturers – VoiceDSP Digital Speech Processor with Master/Slave, Full-Duplex Speakerphone, Multiple Flash and ARAM/DRAM Support

ISD-T360SB
2—SOFTWARE
2.3.1 VCD (VOICE COMPRESSION AND
DECOMPRESSION)
The VoiceDSP processor implements a state of
the art VCD algorithm of the CELP family. The al-
gorithm provides 3 compression rates that can
be selected dynamically (actually, the algorithm
supports more compression rates). PCM record-
ing (no compression) is also provided.
The lowest compression rate of 5.3 Kbit/s enables
about 30 minutes of recording on an 8-Mbit de-
vice (depending on the relative silence period).
The mid-quality compression rate of 9.9 Kbit/s
provides about 16 minutes of voice recording
time. The highest compression rate of 16.8 Kbit/s,
the highest quality recording, stores up to 10 min-
utes on a 8-Mbit device. For detailed information
about recording times refer to table 1-5.
Before recording each message, the microcon-
troller selects one of the three compression rates,
or PCM recording, with the compression_rate
parameter of the R (Record) command. During
message playback the VoiceDSP processor
reads this one byte parameter and selects the
appropriate speech decompression algorithm.
IVS vocabularies can be prepared in either of
the three compression rates, or in PCM format,
using the IVS tool. All messages in a single vocab-
ulary must be recorded using the same algo-
rithm. (See the IVS User’s Guide for more details).
During speech synthesis, the VoiceDSP processor
automatically selects the appropriate speech
decompression algorithm.
SILENCE COMPRESSION
A Voice Activity Detector (VAD) is used in order
to detect periods of silence during the compres-
sion of the recorded message. Silence is treated
differently than normal voice by the compres-
sion algorithm. It is compressed to about 1.0 Kbit/
s. The compressed silence contains data that al-
lows to generate comfort noise during message
playback. The comfort noise generation is impor-
tant because the human ear is not used to “real”
silence while listening to messages.
Various tunable parameters are available in or-
der to optimally tune the VAD. The silence com-
pression may be turned Off, though it is planned
to remain On continuously. For more details refer
to table 2-4 of the Command Description sec-
tion.
NOTE
The silence compression should be turned
Off when ARAM devices are used for voice
storage. Otherwise, unpredictable results
are expected during message playback.
SW AGC
A SoftWare Automatic Gain Control (SW AGC)
algorithm is activated with the compression
module in order to regulate the input signal to a
dynamic range that will provide higher compres-
sion quality. The algorithm senses the energy lev-
el and updates the signal gain in order to amplify
low energy signals and to avoid signal satura-
tion. The SW AGC feature eliminates the need for
an external HW AGC, thus reducing hardware
costs and complexity. Hardware Gain Control
may be used to avoid signal saturation prior to
sampling the signal.
A tunable parameter determines the maximum
allowed gain for the SW AGC algorithm. The SW
AGC may be turned Off, though it is planned to
remain On continuously. For more details refer to
table 2-4 of the Command Description section.
VARIABLE SPEED PLAYBACK
This feature increases or decreases the speed of
messages and synthesized messages during
playback. Use the SPS (Set Playback Speed) to
set the speed of message playback. The new
speed applies to all recorded messages and syn-
thesized messages (only if synthesized using IVS),
until changed by another SPS command. If this
command is issued while the VoiceDSP proces-
sor is in the PLAY state, the speed also changes
for the message currently being played.
The speedup / slowdown algorithm is designed
to maintain the pitch of the original speech. This
approach provides the same speech tone while
playback speed varies.
2-10
Voice Solutions in Silicon^™