Background

In the past, using both hands to grab an infrared remote control, or walking to the device to physically touch buttons to control household devices such as air conditioners, TVs, hoods, and fans may not feel smart or convenient for users. Especially when unable to free both hands or unwilling to move much, a new control method is needed when controlling household appliances -> voice control.

For the convenience of controlling home appliances just by speaking, perhaps only a few simple and commonly used voice command words are needed, easy to remember and control. Users tend to use offline voice control to control home appliances to avoid the hassle of connecting to Wi-Fi, enabling quick responses to control devices, while ensuring personal privacy and security.

The offline voice control solution possesses cost-effectiveness, allowing users to control household appliances using voice recognition at a distance of 5 meters, supporting the integration of 20 to 150 voice command phrases such as 'turn on the lights,' 'set to 25 degrees,' enabling convenient voice control. Using offline voice control makes life more convenient.

Solution Overview

The Unisound single-microphone offline language solution is currently the most cost-effective solution on the market, with built-in offline language recognition models and simple noise reduction front-end algorithms, achieving up to a 95% recognition rate in a 5-meter home environment. Its single-chip solution makes peripheral connections simple. The main control platform operates at a speed of 240MHz, integrating an AI vector processing unit capable of running language recognition models of up to 1M.

In the field of language recognition, this solution can rapidly customize offline language command words, supporting up to 150 command words and 10 wake-up words. It also utilizes MCU resources for simple logic control including UART, GPIO, PWM, aligning solutions rapidly with user needs.

Due to sufficient computational power in the main control chip, noise reduction and AGC (Automatic Gain Control) algorithms are implemented in front of the recognition model. The Unisound single-microphone offline voice solution can pick up sound from a 5-meter distance, achieving a recognition rate of over 90% as long as the voice is 2dB higher than the ambient noise.

Algorithm Architecture

The software architecture of the Unisound offline voice control solution is depicted in the following figure, with the main control being US516P6, featuring two cores for DSP algorithm processing and MCU logic control. The software architecture includes the following processes: - NS algorithm for front-end noise reduction - AGC algorithm for front-end automatic gain adjustment allowing long-distance sound pickup - Model recognition for recognizing voice command phrases - MCU control for executing commands and performing simple GPIO control based on voice recognition

Hardware Design

The Unisound high-performance single-microphone offline voice control solution uses the main control US516P6 as a single-chip solution with a relatively simple circuitry shown in the figure below.

Here: - The 24MHz passive crystal oscillator serves as the oscillation circuit for US516P6 - A simple D-class amplifier is included for speaker playback

Solution Features

Main Control Chip

• US516P6, single-chip solution
• TSOP-28 package
• ~300mW power consumption

Audio Interface

• Supports single-channel analog microphone
• Supports analog Line-out for playback

Audio Algorithms

• NS Noise Suppression
• AGC Long-distance Sound Pickup
• Voice Recognition Template

Application Scenarios

The Unisound offline voice control is suitable for smart homes, air conditioners, hoods, fans, clothes drying racks, and also fits well in 86 panels, remote controls, air conditioning companions, and desk lamps among other individual products.