Offline speech recognition solution development

2020-12-11 17:16:34

Introduction to offline speech recognition solution development


ASR (Automatic Speech Recognition) automatic speech recognition technology is based on key word list recognition technology. In each recognition process, the voice content spoken by the user is converted into voice features through the frequency spectrum, and the items in the keyword list are matched one by one, and the best match is used as the recognition result. For example, in a mobile phone application, the content of this keyword list is the name of the person in the phone book/the menu command of the phone/the name of the song in the T card, etc. No matter what the content of this list is, the user only needs to set the relevant register. The content of the corresponding item to be recognized can be passed to the recognition engine in the form of characters.

The recognition engine does not care about the content of the keywords in the keyword list, which can be any Chinese character strings such as commands, names of people, names of songs, operation instructions, etc.

The maximum number of words supported by each keyword is limited to 30 words from an algorithm perspective. However, from the actual situation, when users say more than 8 words in one breath, they will almost certainly appear typo/missing/speaking multiple words/hiccups/pauses, etc. These situations will seriously affect recognition and cause recognition. error. Therefore, in general, if you want to obtain an ideal recognition effect, it is recommended that the number of words of each keyword is not too long to avoid affecting the effect.


The most important significance of ASR technology is to provide a voice-based user interface that is free of buttons and mice, making the user's operation of the product more natural, faster and more convenient than manual control, and can be used for smart home appliances, smart wear, and voice-activated smart toys. And many other fields.

At present, the price of online speech recognition solutions is relatively high, the product access period is long, and the response time is affected by the network. This solution provides offline voice recognition of non-specific human voice based on a large number of voice databases. This solution has cost advantages, simple access, short cycle, fast identification response speed (within 0.5 seconds), and no need to connect to the network. The overall solution is mature, stable and reliable, allowing traditional products to quickly upgrade the intelligent voice function.


Offline speech recognition technology is a technology based on keyword list recognition. Only need to set up a list of key words to be recognized, and the key words spoken by the user can be recognized through the speech recognition system, without the need for any recording training by the user.

Easy without internet connection

Realize voice control


Market application:

Smart voice lights, smart toys, home appliances (electric fans, range hoods, air conditioners), eye massagers, neck massagers, smart speakers and many other products.


Solution advantages:

Recognition rate is over 95%, accurate recognition within 3M distance

Support 32 languages and multiple minor languages

No need to connect to the Internet, offline identification, quick response

Support voice feedback broadcast, increase interactive entertainment

High cost performance, SOC architecture, integrated Codec

High stability, industrial grade chip

Flexible instruction design, no need for customer recording

Short development time and technical support in place


Technical Parameters:

Operating voltage range is optional 3.3~5.5V

Input dedicated MIC pin, enhanced level is 26dB, maximum total gain is 65dB

Differential audio PWM output, directly drive 1W/8Ω speakers

2 PWM signal output, can be used for LED dimming

Provide I2C/UART interface and custom communication protocol, which can be used for external modules and communication with external MCU



cooperation method:

Provide a complete design plan for voice recognition, customers can design functions according to their own needs, our company is responsible for the development of the voice control part, and docking with the customer's main control

Chip authorization: Customers develop applications themselves, produce PCBs, and purchase corresponding components for production. Our company provides chips, related reference routes, and voice recognition commands.


success case:

1. Intelligent voice light

2. Smart home appliances

3. Health massager

4. Intelligent central control


Voice noise reduction module:

Effectively suppress environmental noise and improve the recognition efficiency in noisy environments.

Suitable for products with mechanical movement, such as fans, heaters, range hoods, sweeping robots, etc.



Welcome all development engineers to come to consult and understand.