Using MCU to implement speech recognition?

灞波儿奔

Using MCU to implement speech recognition? [Copy link]

In a world where everything is connected, more and more electronic devices that can understand speech are coming into our sight. In devices such as smartphones, tablets and laptops with Siri or Cortana applications, speech recognition can help users search for answers or control the electronic devices around them. Although these applications are eye-catching, they take up a lot of processing power and memory. So it’s no surprise that people misunderstand that microcontrollers (MCUs) are too small to recognize speech.

[color=rgb(85, 85, Yes, the low power and small size of an MCU doesn’t make it enough to understand all speech, but for small, low-power embedded applications, recognizing a few well-defined phrases, such as “heat my coffee” or “turn off the lights,” may be enough. Texas Instruments (TI) recently demonstrated this capability on its low-power MSP432 MCUs at CES. TI also released a speech recognizer library written in C that enables MSP432 MCU-based applications to recognize up to 11 phrases that a user frequently uses, ignoring other speech content. Although the user must repeat the phrase they want to be recognized several times during setup, once this step is successfully completed, it can be used smoothly afterwards.

This speech recognizer library has many exciting features, such as:

• Users can change their phrases at any time
• Users can set the recognizer to respond to only a few phrases
• Of course, users can also delete certain phrases in the phrase library according to personal needs

In addition, some easy-to-use header files and user and API guides are provided together with the speech recognition library to help users get started quickly. The download also includes example demonstration programs for the MSP432 MCU LaunchPad development kit, the Audio BoosterPack plug-in module, and the Sharp or Kentec LCD BoosterPack kits. This demonstration application uses the 14-bit analog-to-digital converter (ADC14) integrated in the MSP432 MCU to collect voice and a liquid crystal display (LCD) to display the menu. This menu will allow you to successfully run the recognition function, and the user can choose to:

• First, say a phrase that you want the recognizer to remember. The recognizer will create a model for this speech and store it in flash memory (a task called enrollment).
• Say the enrolled phrase again. The recognizer will use it to make a better model to improve performance (a task called updating).
• Delete the first model that was enrolled
• Run the recognition function