Specifications

79
PF1188-02
Voice recognition by phoneme model
● Japanese speech recognition middleware
for the S1C33 Family
● Real-time execution in an S1C33 chip
● Phoneme model supported
■ FEATURES
• Speech recognition middleware for the S1C33 Family is provided in the form of a linkable library.
• The speech recognition engine and a listening routines using an A/D converter are included in the library.
Speaking routines with a D/A converter or PWM can be implemented.
•Using the Seiko Epson exclusive phoneme model recognition technology, about 100 words can be recognized
in real-time.
• Capable of selecting between two kinds of probability data for phonemes 33 and 85, according to the recogni-
tion rate and memory used.
• Supports Japanese for recognition.
• Voice recognition data can be created simply by just writing words to be recognized in text format.
• PC tools for evaluating the speech recognition engine are provided.
• Evaluation boards are provided.
• Suitable for PDAs, digital stationeries and digital toys.
■ HARDWARE CONFIGURATION
S1C33xxx
S1C33000
Core CPU
Built-in RAM 16-bit timer
External RAM External ROM
Bus interface
10-bit A/D
converter
D/A converter
or PWM
MIC
AMP
Mic
Speaker
POWER
AMP
In case use speak
■ MEMORY REQUIREMENTS AND SPEED
• RAM ................................................................................ approx. 5K bytes
• ROM
Library (Program) ......................................................... approx. 55K bytes
Recognition dictionary.................................................. Four bytes per vowel, 8 bytes per consonant
Probability data for phoneme 33 .................................. approx. 200K bytes
Probability data for phoneme 85 .................................. approx. 523K bytes
• Recognition speed
40MHz operation, recognition dictionary: 50 words ..... 1 to 1.4 seconds
(The dictionary is configured with 5 to 10 phonemes per single entry such as "Good morning" or "Good
afternoon.")
S1C33 Family S5U1C330H1S Middleware