TDA7550

ASR311 Automatic Speech Recognition solution

The TDA-7550 is a single-chip solution for isolated word speech recognition, featuring the Lernout &

Hauspie ASR311 Automatic Speech Recognition engine.

Word base recognition

Isolated word recognition

The words database is created from an extensive set of recordings,

with an equal distribution of speakers; the recordings are partially

taken in automotive conditions at low, medium and high speed.

Quasi-connected digit recognition

Minimum pause between two numbers 150ms.

Noise-robust recognition

The recognition is still very robust even in high noisy environment (as

automotive or industrial environments)

Speaker-independent recognition

Recognition is affected by selected language, no training is required

Speaker-dependent recognition

User words can be stored and mixed with speaker-independent words

Training phase

User words can be trained by repeating three times the selected word.

Recognition is guaranteed when the same speaker is talking.

Prompts

Voice prompts can be stored on FLASH for creating a voice-based

user interface

Control through I

Management of the recognition engine, user words and voice prompts

is accomplished through an I

slave I

Many languages available

Vocabularies in the following languages are available:

US English, French, German, Italian, Spanish, Japanese.

All the vocabularies have a common subset of about 150 words. Other

words and languages are available on request.

The external FLASH is used to store:

♦ Speaker Independent vocabulary (4 KB/word)

♦ User Words (4 KB/word)

♦ Voice prompts (11 KB/sec)

Sample rate:

11.025 kHz

Recognition rate:

> 95%

Maximum number of active words:

30

Maximum number of words:

450

Word memory requirements:

4 KB (speaker-independent)

4 KB (speaker-dependent)

Prompt memory requirements:

11 KB/s

audio amp

spkr

mic preamp

T

0

F

H

b

k

µC

slave I

EMI

ADC

DAC

Up to

4 MB

service request

Figure 1.

TDA7550R - TDA7550 - TDA7551 - TDA7552 - TDA7553

8/12