NXP Semiconductors SLN-LOCAL2-IOT, User Manual

Page: 42 / 87

Figure 48. Inference engine instances of single-language two-stage scenario
The SLN-LOCAL2-IOT kit plays the “can I help you?” and “OK” audio responses respectively when the wake word and commands
are detected. The audio playback files are saved in the filesystem. You can replace them with your own files. For details on the
filesystem, see
Filesystem . For the full list of audio file addresses, see Table 5 .
7.1.1.2 Scenario #2: Multiple-language two-stage voice control
Figure 49. Multiple (up to four) languages of wake word and command inference engines
The SLN-LOCAL2-IOT kit with the i.MX RT106S MCU supports up to four languages of the wake word and command engine
instances, as shown in
Figure 49 . This scenario is a two-stage voice control application. The four language instances for both the
wake word and the command are saved in the flash memory. Users can select any combination of the four language instances.
The selected languages’ wake word engines are loaded into their dedicated RAM memory pool and start receiving the voice data
stream. When one of the wake words is detected, the same language’s command engine instance is loaded into its memory pool
to start listening to the user’s voice command. For example, suppose that two languages (English and Mandarin) are enabled.
The SLN-LOCAL2-IOT kit loads the wake word engines (that is English and Mandarin) into their RAM memory pools and starts
listening to the user’s voice. If the user utters the English wake word “Hey, NXP”, the SLN-LOCAL2-IOT kit detects the wake word,
loads the command engine for English into the RAM memory pool and starts listening to voice commands.
For multiple wake engines listening to the voice stream simultaneously, the False Acceptance Rate (FAR) can
increase. The wake word inference engines must be fine-tuned to mitigate the FAR.
NOTE
It is also possible to load one wake word inference engine, followed by command engines of multiple languages. In this case, the
FAR can be low. Developers must avoid similar pronounciation among different languages’ voice commands.
By default, NXP has implemented voice control demos for the Smart Home, Elevator, Audio Device Control, and Washing Machine
applications in English, Chinese, German, and French. All the available wake words and commands are listed in
Table 7 .
NXP Semiconductors
Far-field local voice control framework
SLN-LOCAL2-IOT Developer’s Guide, Rev. 0, 19 April 2021
User's Guide 42 / 87