Sound recognition system and method based on FPGA platform
A sound recognition and platform technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problems of complex detection of ambient sound, and achieve the effect of increasing the available range, saving equipment costs, and reducing energy consumption
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0043] Such as figure 1 Shown, in one embodiment of the present invention, a kind of sound recognition system based on FPGA platform comprises interconnected FPGA development board and sound input module;
[0044] Wherein, the FPGA development board includes sound input preprocessing module, sound MFSC extraction module, CNN recognition module, DMA (direct memory access) module, DDR3, DDR Ctrl, data cache deployment control module, ping-pong RAM, MFSC RAM, CNN input RAM and CNN output RAM; the DMA module also includes a first central direct memory submodule (CDMA_0), a second central direct memory submodule (CDMA_1), a third central direct memory submodule (CDMA_2) and a fourth central direct memory submodule module(CDMA_3);
[0045] Described sound input module is connected with described data cache control module through described sound input preprocessing module, ping-pong RAM, sound MFSC extraction module and MFSC RAM successively; Described data cache control module is a...
Embodiment 2
[0055] This embodiment is aimed at the specific structure of the sound input preprocessing module, such as figure 2 As shown, the sound input preprocessing module includes a CTRL submodule, an ADC_READ submodule and a FIR submodule, wherein the CTRL submodule is connected to the ADC_READ submodule and the FIR submodule respectively, and the ADC_READ submodule is also connected to the FIR submodule The FIR sub-module is connected; the FIR sub-module is also connected with the ping-pong RAM.
[0056] The CTRL submodule includes IIC_CTRL, FIR_CTRL and RAM_CTRL connected in sequence, the RAM_CTRL is also connected to the ping-pong RAM, the FIR_CTRL is also connected to the FIR submodule, and the IIC_CTRL is also connected to the sound input module.
[0057] The CTRL (internal control) sub-module includes IIC_CTRL, FIR_CTRL and RAM_CTRL, among which IIC_CTRL is used for the configuration of the WM8731 chip, and is also used to notify the ADC_READ sub-module to receive the data of ...
Embodiment 3
[0061] The present embodiment is aimed at the concrete structure of sound MFSC extraction module, as image 3 Shown, described sound MFSC extraction module comprises the sound MFSC extraction module state machine that connects successively, add window submodule (add_win), Fourier leaf module (fft), amplitude calculation submodule (abs_calc), Mel filter submodule ( mel_fil), natural logarithm calculation submodule (ln_calc) and the first filling submodule (log_mel_pad); the sound MFSC extraction module state machine is also connected with the ping-pong RAM, and the first filling submodule is also connected with the MFSC RAM connection;
[0062] The first filling sub-module is used to fill the output MFSC data. The MFSC data can be regarded as a feature map. The filling method is to fill 1 row or 1 column of 0 data around the feature map to meet the requirements of the CNN recognition module. Input data requirements.
[0063] The Mel filtering sub-module includes a Mel filter ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


