09-10-2012, 10:18 AM
SPEECH RECOGNITION MICROCONTROLLER
[attachment=35709]
The RSC-464 is the newest member of Sensory’s RSC-4x family of speech microcontrollers. The RSC-464 has
many features of the RSC-4128, but costs less due to reduced integrated memory.
The RSC-464 is designed to bring high performance speech I/O features to the most cost-sensitive embedded
consumer products. The RSC-464 operates in tandem with the radically new FluentChip™ technology, offering
the best embedded speech recognition technologies in the industry. FluentChip is capable of accurate speech
recognition; voice password; high quality, low data-rate compressed speech; and music synthesis. One or more
technologies may be used in an application. Revolutionary Text-to-Speaker-Independent (T2SITM) technology is
derived from a Hidden Markov Model-Neural Net hybrid, allowing the creation of SI recognition sets in minutes
by simply entering text.
The RSC-464 is based on an 8-bit microcontroller and integrates speech-optimized digital and analog blocks
into a single chip solution. The RSC-464 can store technology code, data sets, and application code on-chip as
a single chip solution. The RSC-464 acts as a general purpose controller, providing the option of replacing an
existing microcontroller for cost optimal solutions, or by alternatively acting as a speech recognition slave.
In addition to best-in-class embedded speech recognition performance, the RSC-464 provides further on-chip
integration of features. A complete speech I/O application can be built with as few additional parts as a clock
crystal, speaker, microphone, and few resistors and capacitors.
The RSC-464 is designed to operate in tandem with the FluentChip™ technology library, including speaker
independent (SI), speaker dependent (SD), and speaker verification (SV) speech recognition. Combinations of
these technologies may used to create applications that are rich in features. These are described below:
Speaker Independent recognition requires no user training. The RSC-464 can recognize up to 10 words
in an active set (number of sets is limited only by internal ROM size). Text-to-SI (T2SI), based on a
hybrid of Hidden Markov Modeling and Neural Net technologies, allows creation of accurate SI
recognition sets in seconds. SI requires on-chip ROM to store the words to be recognized.
Speaker Dependent recognition allows the user to create names for products or customize recognition
sets. SD is implemented with DTW (dynamic time warping) pattern matching technology. SD requires
programmable memory to store the personalized speech templates that may be on-chip SRAM or offchip
Serial EEPROM, serial Flash Memory, or serial SRAM. Up to 30 words can be recognized in an
active set (the number of unique sets is limited only by programmable memory capacity). The RSC-464
can store up to 2 SD templates (trained patterns) in on-chip SRAM.
Speaker Verification enables the RSC-464 to authenticate when a previously trained password is
spoken by the target user. SV is also implemented with DTW technology. Two SV templates can be
stored in on-chip SRAM, or more with external programmable serial memory.
Word Spotting enables the RSC-464 to spot a specific word surrounded by other speech within a
phrase. This can be quite effective when the users response may vary (e.g. spotting “telephone” in the
phrases “ummm telephone”, or “telephone call”). This option is available for SI and SD.
Continuous Listening allows the chip to continuously listen for a specific word. This may be used as a
trigger word to request a device to listen for a command. This option is available for SI and SD.
[attachment=35709]
The RSC-464 is the newest member of Sensory’s RSC-4x family of speech microcontrollers. The RSC-464 has
many features of the RSC-4128, but costs less due to reduced integrated memory.
The RSC-464 is designed to bring high performance speech I/O features to the most cost-sensitive embedded
consumer products. The RSC-464 operates in tandem with the radically new FluentChip™ technology, offering
the best embedded speech recognition technologies in the industry. FluentChip is capable of accurate speech
recognition; voice password; high quality, low data-rate compressed speech; and music synthesis. One or more
technologies may be used in an application. Revolutionary Text-to-Speaker-Independent (T2SITM) technology is
derived from a Hidden Markov Model-Neural Net hybrid, allowing the creation of SI recognition sets in minutes
by simply entering text.
The RSC-464 is based on an 8-bit microcontroller and integrates speech-optimized digital and analog blocks
into a single chip solution. The RSC-464 can store technology code, data sets, and application code on-chip as
a single chip solution. The RSC-464 acts as a general purpose controller, providing the option of replacing an
existing microcontroller for cost optimal solutions, or by alternatively acting as a speech recognition slave.
In addition to best-in-class embedded speech recognition performance, the RSC-464 provides further on-chip
integration of features. A complete speech I/O application can be built with as few additional parts as a clock
crystal, speaker, microphone, and few resistors and capacitors.
The RSC-464 is designed to operate in tandem with the FluentChip™ technology library, including speaker
independent (SI), speaker dependent (SD), and speaker verification (SV) speech recognition. Combinations of
these technologies may used to create applications that are rich in features. These are described below:
Speaker Independent recognition requires no user training. The RSC-464 can recognize up to 10 words
in an active set (number of sets is limited only by internal ROM size). Text-to-SI (T2SI), based on a
hybrid of Hidden Markov Modeling and Neural Net technologies, allows creation of accurate SI
recognition sets in seconds. SI requires on-chip ROM to store the words to be recognized.
Speaker Dependent recognition allows the user to create names for products or customize recognition
sets. SD is implemented with DTW (dynamic time warping) pattern matching technology. SD requires
programmable memory to store the personalized speech templates that may be on-chip SRAM or offchip
Serial EEPROM, serial Flash Memory, or serial SRAM. Up to 30 words can be recognized in an
active set (the number of unique sets is limited only by programmable memory capacity). The RSC-464
can store up to 2 SD templates (trained patterns) in on-chip SRAM.
Speaker Verification enables the RSC-464 to authenticate when a previously trained password is
spoken by the target user. SV is also implemented with DTW technology. Two SV templates can be
stored in on-chip SRAM, or more with external programmable serial memory.
Word Spotting enables the RSC-464 to spot a specific word surrounded by other speech within a
phrase. This can be quite effective when the users response may vary (e.g. spotting “telephone” in the
phrases “ummm telephone”, or “telephone call”). This option is available for SI and SD.
Continuous Listening allows the chip to continuously listen for a specific word. This may be used as a
trigger word to request a device to listen for a command. This option is available for SI and SD.