DSP turns microphone array into steerable listening beam for smart control
- 作者：Ella Cai
- 发布时间：：2017-06-13
Bristol-based XMOS has created a reference design for home listening devices, like Amazon’s Echo, that work with voice recognition algorithms.
Based around the firm’s digital processing silicon, it converts an array of up to four microphones into a single far-field microphone with a range of 5m or more, XMOS v-p marketing Paul Neil told Electronics Weekly.
It includes per-microphone acoustic echo cancellation, then adaptive beam-forming to generate a single audio signal from the microphone array, followed by dynamic echo and noise suppression, and automatic gain control.
Two outputs are produced – one best suited to automatic speech recognition, and one optimised for people – for a speakerphones, for example. “The human ear doesn’t like the non-linearities introduced by the processing signal chain,” said Neil.
XMOS prototype smart speakerTwo sets of reference hardware are available, one with a circular microphone array for 360deg listening (pictured) and one with a linear array for 180deg.
On top of this, via intellectual property from Sensory, there is optional key-word recognition – which can provide similar function to the “Alexa” wake-up in Amazon’s system.
The reference design has been dubbed XVR3000 (XVR310x with Sensory), and is the first in a series of acoustic-related reference designs from XMOS, said Neil. Branded ‘VocalFusion’ they will include: loudspeakers, TV soundbars, computer conferencing, domestic appliance control and robotic interfacing.
XMOS’s silicon is unique in the market, each device consisting of a different number of identical cores that can be programmed as processing units or peripherals. Performance is fast, allowing peripherals such as I2C and USB interfaces to be implemented in software – much of the peripheral software is available from the company.
As such, XVR3000/310x (which uses 16 core silicon) could be added as-is to an existing customer Wi-Fi router/hub design, said Neil, or the functions of the router, or other host hardware, can also be implemented on XMOS cores. “You can imaging running Wi-Fi stacks on XMOS as well, creating an XMOS-based voice thin-client to interface with Amazon or Google,” he added.
Based around the firm’s digital processing silicon, it converts an array of up to four microphones into a single far-field microphone with a range of 5m or more, XMOS v-p marketing Paul Neil told Electronics Weekly.
It includes per-microphone acoustic echo cancellation, then adaptive beam-forming to generate a single audio signal from the microphone array, followed by dynamic echo and noise suppression, and automatic gain control.
Two outputs are produced – one best suited to automatic speech recognition, and one optimised for people – for a speakerphones, for example. “The human ear doesn’t like the non-linearities introduced by the processing signal chain,” said Neil.
XMOS prototype smart speakerTwo sets of reference hardware are available, one with a circular microphone array for 360deg listening (pictured) and one with a linear array for 180deg.
On top of this, via intellectual property from Sensory, there is optional key-word recognition – which can provide similar function to the “Alexa” wake-up in Amazon’s system.
The reference design has been dubbed XVR3000 (XVR310x with Sensory), and is the first in a series of acoustic-related reference designs from XMOS, said Neil. Branded ‘VocalFusion’ they will include: loudspeakers, TV soundbars, computer conferencing, domestic appliance control and robotic interfacing.
XMOS’s silicon is unique in the market, each device consisting of a different number of identical cores that can be programmed as processing units or peripherals. Performance is fast, allowing peripherals such as I2C and USB interfaces to be implemented in software – much of the peripheral software is available from the company.
As such, XVR3000/310x (which uses 16 core silicon) could be added as-is to an existing customer Wi-Fi router/hub design, said Neil, or the functions of the router, or other host hardware, can also be implemented on XMOS cores. “You can imaging running Wi-Fi stacks on XMOS as well, creating an XMOS-based voice thin-client to interface with Amazon or Google,” he added.
