NXP Semiconductors SLN-LOCAL2-IOT, User Manual

Page: 41 / 87

7.1.1 ASR application scenarios
Figure 47 shows the ASR with multiple languages, where each language consists of a wake word inference engine and
N
instances of command inference engines. Developers can implement various application scenarios that are described in the
following subsections.
The SLN-LOCAL2-IOT with i.MX RT106S supports up to four languages of the ASR in runtime. With i.MX RT105S, it reduces to
two languages because of the reduced RAM size.
Figure 47. Inference engine instances matrix for flexible ASR applications
The simplest ASR application is a single-language two-stage ASR with only one wake word engine and one command engine.
The ASR for multiple (up to four) languages can be created with the inference engine instances of
columns
in Figure 47 . The ASR
for the multiturn (e.g. dialog) application can be created with the inference engine instances of a
row
in Figure 47 .
7.1.1.1 Scenario #1: Single-language two-stage voice control
Figure 48 shows the simplest ASR scenario with one wake word inference engine followed by a command engine instance in a
selected language. By default, NXP has implemented the
LED voice control demo in English
language.
• Wake word
— Hey, NXP
• Commands
— L, E, D, red
— L, E, D, green
— L, E, D, blue
— Cycle fast
— Cycle slow
NXP Semiconductors
Far-field local voice control framework
SLN-LOCAL2-IOT Developer’s Guide, Rev. 0, 19 April 2021
User's Guide 41 / 87