NLP series AI processor

Descriptions 

Analog input channels: 12

Analog output channels: 8

AI noise reduction: The AI noise reduction function based on width learning can intelligently eliminate environmental noise such as turning books, writing, knocking on tables, fans, etc., and achieve non-human voice signal separation;

Width learning (100,000 hours): The network structure continues to change with learning, and the number of network nodes can be dynamically adjusted, making it easy to implement online learning with extremely low latency;

Processor: ADI SHARC 21489@450 MHz SIMD;

DSP processing capability: 400 MIPS, 1.6 GFLOPS;

Sampling rate: 48 kHz, ± 100 ppm;

THD+N:<-95dB @17dBu

Input dynamic range: 113dB

Output dynamic range: 113dB

Provides 8 devices for online operation at the same time;

Built-in one-in and one-out USB sound card, supports music playback, recording and soft video conferencing (such as ZOOM, Tencent Conference, DingTalk Conference, etc.);

Bus-type AEC, tail time: 512ms, convergence rate: 60dB/S, echo cancellation amplitude: 60dB;

Independent channel AFC (feedback suppression), using notch algorithm, sound transmission gain improvement: 10dB;

Noise suppression (ANS), signal-to-noise ratio increased by 18dB

Gain sharing automatic mixing (AMC),

8-band British parametric equalizer, providing 5 filter options: Parametric, Lowshelf, Highshelf, Lowpass, Highpass;

Provide end-user customized operation interface, supporting up to 30 devices managed by the same interface;

Support camera tracking;

With 8 channels of GPIO;

It has a central control function and supports three control methods: RS232, RS485, and UDP. It can control the power supply, signal switching, environment control, audio, etc. in the system as a whole, and realize the functions required to start the system with one click.

Features 

A customized user interface needs to be provided;

Requires built-in USB sound card to support recording and remote conferencing;

The processor chip adopts ADI architecture, no less than 40bit DSP floating point operation engine, and provides an open architecture;

AI noise reduction: The AI noise reduction function based on width learning can intelligently eliminate environmental noise such as turning books, writing, knocking on tables, fans, etc., achieving non-human voice signal separation;

Width learning (100,000 hours): The network structure evolves with learning, dynamically adjusting node numbers for ultra-low latency online updates;

Third-party device control integration;

Analog I/O channels: not less than 12*8;

Input/output quantization: not less than 48KHz/24bit;

Input channels include:

- Input gain adjustable in 3dB steps, no less than 16 levels;

- No less than 8-band PEQ with five filter types;

- Adaptive Echo Cancellation (AEC), Noise Suppression (ANS);

- Gain sharing automatic mixing (AMC), Automatic gain (AGC);

- Camera tracking;

- 16-point AFC per channel;

Output channels feature:

- No less than 8-band PEQ, frequency dividing, delay, limiting;

- 16x9 matrix minimum;

- 16 preset configurations;

- Channel LINK and grouping supported;

Dynamic range (Input/Output): not less than 113dB;

Frequency response: not less than 20Hz~20kHz (±0.3dB);

THD+N: not less than -95dB @17dBu.