Intelligent speech recognition system and method

A technology of intelligent speech and recognition system, applied in the field of electronics, can solve the problems of small calculation amount, large computing power consumption and cost, large computing resources, etc. Effect

Active Publication Date: 2016-09-28
成都启英泰伦科技有限公司
View PDF4 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Due to the huge amount of calculation of the DNN model, the current speech recognition calculation is mainly carried out through the cloud server. Huge, and the power consumption is also very high, and it also needs to consume a huge network bandwidth, and the recognition calculation cannot be performed when there is no network or the network is disconnected
If a system-on-a-chip (SoC) integrating single-core or multi-core high-performance CPU or GPU and digital signal processing (DSP) core is used locally for calculation, only small DNN model calculations can be performed, or due to calculation power consumption and The cost is huge, which has a great impact on the competitiveness of end products
[0003] There are also some chips on the market that use non-DNN model methods such as the Gaussian mixture model method. Although this method has a small amount of calculation, it can be undertaken by ordinary CPU cores, but the recognition performance is much worse than the speech recognition method based on the DNN model.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Intelligent speech recognition system and method
  • Intelligent speech recognition system and method
  • Intelligent speech recognition system and method

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example

[0029] Such as figure 2 As shown, an intelligent speech recognition system includes a signal receiving conversion unit 101, INPUT RAM (input buffer) 1021, VAD (voice detection unit) 103, FE (voice feature extraction unit) 104, Feature RAM (feature buffer) 1022, DNN (Deep Neural Network Unit) 105, DNN RAM (Deep Neural Network Cache) 1023, CPU (Central Processing Unit) 106 and FLASH (Flash Memory) 107, the signal receiving and converting unit 101 converts the external sound input signal into a digital signal in a unified format , the digital signal in this embodiment is a PCM (Pulse Code Modulation) signal, and the converted digital signal is buffered into the INPUT RAM1021, and the VAD103 reads the digital signal from the INPUT RAM1021 for voice detection, and detects if there is a voice signal in the digital signal , then VAD103 will send a trigger signal to FE104, DNN105 and CPU106, so that FE104, DNN105 and CPU106 enter the working state, FE104 reads the digital signal from...

no. 2 example

[0032] Such as image 3As shown, an intelligent speech recognition system includes a signal receiving conversion unit 201, INPUT RAM (input buffer) 2021, VAD (voice detection unit) 203, FE (voice feature extraction unit) 204, Feature RAM (feature buffer) 2022, DNN (deep neural network unit) 205, DRAM controller (dynamic random access memory controller) 2023, CPU (central processing unit) 206, FLASH (flash memory) 207 and DRAM (dynamic random access memory) 208, CPU 206 when booting Import the DNN parameters and voice model library pre-stored in the external memory FLASH207 into the DRAM208, and the signal receiving and converting unit 201 converts the external sound input signal into a digital signal of a unified format. The digital signal in this embodiment is a PCM (pulse code modulation) signal , the converted digital signal is cached into INPUT RAM2021, and VAD203 reads the digital signal from INPUT RAM2021 for voice detection, and if there is a voice signal in the digital...

no. 3 example

[0035] Such as Figure 5 Shown, a kind of intelligent speech recognition method comprising the intelligent speech recognition system described in the first embodiment, its steps are as follows:

[0036] Step 1 301 The signal receiving conversion unit receives the external sound input signal and converts the external sound input signal into a digital signal in a unified format. In this embodiment, the external sound input signal is a serial digital I2S (integrated circuit built-in audio bus) signal. In other In the embodiment, the external sound input signal can also be a PDM (Pulse Density Modulation) signal, or a mixture of a serial digital I2S signal and a PDM signal, which is converted into a PCM digital signal by the receiving conversion unit;

[0037] In step two 302, the digital signal is stored in the INPUT RAM for other units to read;

[0038] Step 3 303 VAD reads the digital signal from the INPUT RAM and detects whether there is a voice signal. If not, the process is...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention relates to the technical field of electronics, and particularly relates to an intelligent speech recognition system. The intelligent speech recognition system comprises a signal receiving and converting unit, a storage management unit, a speech detection unit, a speech feature extraction unit, a deep neural network unit and a recognizing and decoding unit. The invention further discloses an intelligent speech recognition method containing the intelligent speech recognition system. The system and the method provided by the invention significantly improve the calculation performance of a chip under the condition of the same chip area compared with a CPU or a GPU, and reduce the power consumption and the cost.

Description

technical field [0001] The invention relates to the field of electronic technology, in particular to an intelligent voice recognition system and method. Background technique [0002] With the breakthrough of artificial intelligence algorithm, deep neural network (DNN) has been applied in speech intelligent recognition, which has greatly improved the accuracy of speech recognition, making speech recognition gradually popularized and applied in various fields in the past two years. Due to the huge amount of calculation of the DNN model, the current speech recognition calculation is mainly carried out through the cloud server. It is huge, and the power consumption is extremely high, and it also needs to consume a huge network bandwidth, and the recognition calculation cannot be performed without the network or when the network is disconnected. If a system-on-chip (SoC) integrating single-core or multi-core high-performance CPU or GPU and digital signal processing (DSP) core is...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G10L15/02G10L15/16G10L15/28
CPCG10L15/02G10L15/16G10L15/28Y02D10/00
Inventor 何云鹏
Owner 成都启英泰伦科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap