Skip to content

Command Word Self-Learning Solution Development

1 Solution Overview

1.1 Background and Objectives

China’s vast territory encompasses numerous dialect groups, primarily categorized into seven major types (such as Cantonese, Wu, Xiang, Gan etc.), each further divided into multiple sub-dialects and accents. Due to significant differences in pronunciation, vocabulary, and grammar compared to Mandarin, traditional Mandarin-trained speech recognition systems struggle with accurate recognition. To address this challenge, Chipintelli, as the inventor of BNPU, has developed a Command Word Self-Learning Solution leveraging BNPU’s efficient AI computing capabilities.

1.2 Technical Principle

The solution converts speaker voice features into deep acoustic features through a deep neural network. During the learning phase, these features are recorded as templates. During recognition, the extracted deep acoustic features are compared with the templates using DTW (Dynamic Time Warping) technology for cumulative distance calculation to achieve recognition.

1.3 Learning Workflow

In this solution, Chipintelli’s AI voice chips (CI13XX or CI13LC) initiate the self-learning process through voice commands or other active commands. The workflow is as follows:

Figure 1.3 Command Word Self-Learning Workflow

Reference video: ☞Offline Self-Learning Solution Demo-lighting (Chinese Version)

2 Development Preparation

Before development, please prepare the following hardware and software. For mass production, please refer to ☞Production Testing and ☞Contact Us for official Chipintelli production support:

2.1 Hardware Preparation

2.1.1 This solution recommends using hardware modules D03GS01J or F162GS02J:

2.1.2 Below is the UART programming tool:

Figure 2.1.2 UART Tool Physical Diagram

Note

Developers can use general UART tools for firmware programming and log viewing, or click Purchase Link

Important: For data stability, it is recommended to use a UART with crystal oscillator. We recommend using the CH341 UART debugging tool (Chipintelli AI Voice UART Debugging Tool)

2.2 Software Development Preparation

2.2.1 Register and log in to the AI Development Platform: https://aiplatform.chipintelli.com

2.2.2 Download the SDK development package with self-learning functionality for the corresponding chip: https://aiplatform.chipintelli.com/attachment

  • CI 13XX Algorithm SDK: CI13XX_SDK_ASR_ALG_Vx.x.x
  • CI 13LC Standard SDK: CI13LC_SDK_ASR_Offline_Vx.x.x (This feature is in the cwsl_sample project within the SDK)

3 Solution Development

3.1 Option 1: Voice Recognition Firmware and SDK Development

3.1.1 Log in to the Chipintelli AI Speech Development Platform and navigate to the component.

3.1.2 Select Voice Recognition Firmware and SDK Development.

3.1.3 Create a new project. Here we use the 1302 chip model for testing (change to 13LC if needed).

3.1.4 Create a project. Note that natural speech projects do not include self-learning capabilities.

3.1.5 Click Continue to configure the firmware parameters according to your requirements.