搜索
bottom↓
回复: 57

[分享]麦克风阵列相关资料和开源项目

  [复制链接]

出0入0汤圆

发表于 2016-9-4 10:29:33 | 显示全部楼层 |阅读模式
基本原理:

http://www.labbookpages.co.uk/audio/beamforming.html

IntroductionThe diagram below shows the sensitivity patterns for two different microphone setups. The lefthand diagram shows the pattern for an ideal omnidirectional microphone. It shows that the microphonehas equal sensitivity to signals from all directions. The right hand image shows a focusedsensitivity pattern aiming for maximum sensitivity in a single direction, whilst all other direction havereduced sensitivity, the goal is to create a sensitivity pattern that results in the abilityto 'listen' to signals coming from a single direction.
The beamforming effect can be achieved by using a simple linear array of microphones. Such anarray is illustrated below, in this case the array has three microphones. It is easy to see that thedirection from which a wave front originates has an effect on the time at which the signal meets each element inthe array. When arriving from -45° the signal reaches the left hand microphone first, whenarriving from perpendicular to the array (called broadside) the signal reaches each microphone atthe same time and when from +45° the right hand microphone receives the signal first.
If the array's output is created by summing all microphone signals, the maximum output amplitudeis achieved when the signal originates from a source perpendicular to the array; the signals arriveat the same time, they are highly correlated in time and reinforce each other. Alternatively, if thesignal originates from a non-perpendicular direction, they will arrive at different times so will beless correlated and will result in a lesser output amplitude.

Beam PatternA simple calculation can be used to determine the sensitivity of a microphonearray for signals coming from a particular direction. The image below shows an array with fourmicrophones. Each array is separated by a distance l (in metres). The angle of arrival is measured fromperpendicular to the array. The equation below calculates the array's gain for a single frequencyf, and an arrival angle θ. c denotes the speed of sound andN is the number of microphones.
The page on Wave Summation gives a full explanation on how this equation is derived.
Note: The equation makes a few assumptions; the signal source is sufficiently far from the array thatthe wavefront is effectively flat, also the there is no accounting for attenuation of the signal as it travels from source to the microphones.
The gain of the array is shown in the plots below. The output is normalised to the output thatwould be received from a single microphone. Therefore at an angle of 0 degrees (broadside) theoutput amplitude is equivalent to a omnidirectional microphone, resulting in a gain of 1 (or 0dB).
Beam Pattern, 4 Elements, 0.2m Spacing, 1kHz
Beam Pattern in polar form
The data for these plots was generated using the code below. It was then plotted using gnuplot. Twosets of gnuplot commands are also given below, one for the XY plot, the other for the polar plot.

Frequency ResponseThe sensitivity-plot calculation in the previous section was calculated for a single frequency.When dealing with a broadband source (such as speech) it is important to calculate an array'sfrequency response. The plot below shows the array's frequency response for the frequency range 0 to10000Hz. The most noticeable features are the gain maximas where the array's output is equal to thatat the perpendicular source direction. These are called grating lobes and are explained below. Their effect is a loss of directional filtering. Signal sources arriving fromnon-perpendicular directions will make it to the array's output in bands of frequencies.
Another interesting feature is the lack of directionality at low frequencies.
The code used to generate the data for this plot is show below. The gnuplot commands are also given.

Microphone Spacing & QuantityAn inspection of the delay-sum equation from the Beam Pattern section reveals that the beamformer's performance isdependant on the spacing and quantity of the array's elements. The table below illustrates how changing these parametersaffects the beamformer's spatial filtering performance.

5 Elements, 0.04m Spacing, 0.2m Aperture
15 Elements, 0.04m Spacing, 0.6m Aperture
25 Elements, 0.04m Spacing, 1m Aperture
5 Elements, 0.08m Spacing, 0.4m Aperture
15 Elements, 0.08m Spacing, 1.2m Aperture
25 Elements, 0.08m Spacing, 2m Aperture
5 Elements, 0.16m Spacing, 0.8m Aperture
15 Elements, 0.16m Spacing, 2.4m Aperture
25 Elements, 0.16m Spacing, 4m Aperture

Grating LobesBelow is a beam pattern plot for a 4 element array with 0.2m spacing at a frequency of 4kHz. Theplot shows a number of extra lobes with a gain that matches the main lobe, these are called gratinglobes. Grating lobes are usually unwanted as they cause the array to pick up signals from directionsother than that of the main lobe without attenuation.
Grating lobes occur when the extra distance a signal wavefront must travel between array elementsis a multiple of the signal's wavelength. In this situation, the signals received by each elementare highly correlated and no signal attenuation occurs. This is illustrated in the right hand diagrambelow.
For a 1 dimensional linear array with equally spaced elements the positions of the grating lobesare easy to calculate. l is the element spacing, c is the speed of sound,f is the signal frequency and n is an integer that selects the gratinglobe.
       
nangle (degrees)
00.0°
125.388°
259.037°

SteeringSo far the main lobe of the array's sensitivity pattern has been fixed to a single direction;perpendicular to the array. A powerful feature of the beamforming array is the ability toelectronically steer the beam pattern without physically moving the array. This is simply achieved byadding a delay stage to each of the array elements. The diagram below illustrates this configuration, that givesthe Delay-Sum architecture its name.
The idea is very simple, add a delay to each microphone such that the signals from a particulardirection are aligned before they are summed. By controlling these arrays, the main lobe directioncan be steered.

=====================================================================================================
=====================================================================================================
开源项目:manyears 我就懒得翻译了
ManyEars Microphone Array-Based Audition for Mobile Robots
ManyEars implements real-time microphone array processing to perform sound source localisation, tracking and separation. It was designed for mobile robot audition in dynamic environments. This project was started at IntroLab, Universite de Sherbrooke, Sherbrooke, Quebec, Canada.
Features
  • Real-Time
  • Sound Source Localisation
  • Sound Source Tracking
  • Sound Source Separation
  • Tuning GUI
  • Microphone Array
  • Beamforming

https://github.com/introlab/manyears

本帖子中包含更多资源

您需要 登录 才可以下载或查看,没有帐号?注册

x

阿莫论坛20周年了!感谢大家的支持与爱护!!

如果想吃一顿饺子,就得从冰箱里取出肉,剁馅儿,倒面粉、揉面、醒面,擀成皮儿,下锅……
一整个繁琐流程,就是为了出锅时那一嘴滚烫流油的热饺子。

如果这个过程,禁不住饿,零食下肚了,饺子出锅时也就不香了……《非诚勿扰3》

出0入8汤圆

发表于 2016-9-4 10:33:46 来自手机 | 显示全部楼层
好资料,谢谢楼主分享。

出0入0汤圆

发表于 2016-9-4 11:53:10 | 显示全部楼层
好资料,感谢分享

出0入0汤圆

发表于 2016-11-6 23:16:16 | 显示全部楼层
感谢分享...........

出0入0汤圆

发表于 2016-11-6 23:49:07 | 显示全部楼层
不错的资料,xmos有直接的方案,

出0入0汤圆

发表于 2016-11-7 02:43:59 | 显示全部楼层
阵列这东西还真很有意思.

出0入4汤圆

发表于 2016-11-7 10:32:37 | 显示全部楼层
好资料,感谢分享

出0入0汤圆

发表于 2016-11-7 10:46:35 | 显示全部楼层
这个是好东西!

出0入0汤圆

发表于 2016-11-7 11:26:55 | 显示全部楼层
好东西,收藏一下

出0入0汤圆

发表于 2016-11-7 11:34:36 | 显示全部楼层
楼主有继续下去吗?有具体的项目吗?

出0入0汤圆

 楼主| 发表于 2016-11-7 11:51:13 来自手机 | 显示全部楼层
overheat 发表于 2016-11-7 11:34
楼主有继续下去吗?有具体的项目吗?

在搞,自用……

出0入0汤圆

发表于 2016-11-7 14:17:36 | 显示全部楼层
感谢楼主,研究过一段相似的东西

出0入0汤圆

发表于 2016-11-7 16:24:14 | 显示全部楼层
是不是狙击手定位系统?这东西可算是很NB的哦!要是机器狙击手就无所谓了!

出0入0汤圆

发表于 2016-11-7 16:33:04 | 显示全部楼层
高端。。。顶楼主。

出0入0汤圆

 楼主| 发表于 2016-11-7 17:02:03 来自手机 | 显示全部楼层
activeleo 发表于 2016-11-7 16:24
是不是狙击手定位系统?这东西可算是很NB的哦!要是机器狙击手就无所谓了! ...

呃,我这是智能家居,全房(所有房间)语音识别的采集器

出0入0汤圆

发表于 2016-11-7 17:14:00 | 显示全部楼层
xivisi 发表于 2016-11-7 17:02
呃,我这是智能家居,全房(所有房间)语音识别的采集器

你要是真能定位狙击手,那真的是很厉害了!国内好多人都搞不定,而美军已经装备了!

出0入0汤圆

发表于 2016-11-7 17:32:35 | 显示全部楼层
xivisi 发表于 2016-11-7 17:02
呃,我这是智能家居,全房(所有房间)语音识别的采集器

我们也在搞,现在这个好像挺火啊。你的主控用什么?XMOS似乎似乎比较合适的。
我想用STM32来做,不过资源有点错位,不够用。

出0入4汤圆

发表于 2016-11-7 18:58:24 | 显示全部楼层
高大上!我们也想搞,不过是超声波矩阵发射!

出0入0汤圆

发表于 2016-11-7 18:59:51 | 显示全部楼层
activeleo 发表于 2016-11-7 17:14
你要是真能定位狙击手,那真的是很厉害了!国内好多人都搞不定,而美军已经装备了! ...

  国内停留在玩具水平的

出0入0汤圆

发表于 2016-11-7 19:02:11 | 显示全部楼层
感觉好高深的样子

出0入0汤圆

 楼主| 发表于 2016-11-7 20:34:05 | 显示全部楼层
activeleo 发表于 2016-11-7 17:14
你要是真能定位狙击手,那真的是很厉害了!国内好多人都搞不定,而美军已经装备了! ...

话说  狙击手 定位,开枪后定位?   1千米左右吧,各种建筑物导致的反射,衍射……想想就头痛

出0入0汤圆

 楼主| 发表于 2016-11-7 20:55:23 | 显示全部楼层
overheat 发表于 2016-11-7 17:32
我们也在搞,现在这个好像挺火啊。你的主控用什么?XMOS似乎似乎比较合适的。
我想用STM32来做,不过资源 ...

自做自用,用FPGA做同步采集,后面国产多核A7处理器(便宜),顺便还有其他业务跑在上面。

出0入0汤圆

发表于 2016-11-7 21:02:07 | 显示全部楼层
好厉害 好高端

出0入0汤圆

 楼主| 发表于 2016-11-7 21:04:15 | 显示全部楼层
本帖最后由 xivisi 于 2016-11-7 21:05 编辑
overheat 发表于 2016-11-7 17:32
我们也在搞,现在这个好像挺火啊。你的主控用什么?XMOS似乎似乎比较合适的。
我想用STM32来做,不过资源 ...


再PS:STM32,性能远远不够,除非阵列规模小,只做简易信号处理

出0入0汤圆

发表于 2016-11-7 23:41:37 | 显示全部楼层
好资料,谢谢

出0入0汤圆

发表于 2016-11-8 09:09:37 | 显示全部楼层
高大上的东西,看不懂只会用模块

出0入0汤圆

发表于 2016-11-8 09:27:13 | 显示全部楼层
曾经做过STM32的MEMS 语音定位项目.研究过一段时间的漂过

出0入0汤圆

发表于 2016-11-8 16:41:00 | 显示全部楼层
xivisi 发表于 2016-11-7 21:04
再PS:STM32,性能远远不够,除非阵列规模小,只做简易信号处理

是啊,我想用STM32当个接口芯片,然后把数据搞到PC上处理。这个怎么样?

出0入0汤圆

 楼主| 发表于 2016-11-8 17:27:34 | 显示全部楼层
overheat 发表于 2016-11-8 16:41
是啊,我想用STM32当个接口芯片,然后把数据搞到PC上处理。这个怎么样?

这个很明显可以啊,你这样STM32基本上只做数据传送,一点问题都没有,除非你的阵列规模比较大,数据量超过了STM32的传输带宽

出0入0汤圆

发表于 2016-11-10 00:00:03 | 显示全部楼层
不错不错

出0入0汤圆

发表于 2017-2-4 15:10:15 | 显示全部楼层
标记一下,谢谢分享!

出0入0汤圆

发表于 2017-2-4 15:11:13 | 显示全部楼层
标记一下,谢谢分享!

出0入0汤圆

发表于 2017-2-4 15:23:05 来自手机 | 显示全部楼层
不错的资料,Mark

出0入0汤圆

发表于 2017-2-6 14:14:32 | 显示全部楼层
XMOS的开发板国内有没有哪家做的,进口的要1500刀,而且我要买6块。

出0入0汤圆

发表于 2017-2-6 14:25:50 | 显示全部楼层
成品,现在优惠价600多。。。。某宝搜索“”ReSpeaker 远场麦克风阵列“”

出0入0汤圆

发表于 2017-3-24 12:49:09 | 显示全部楼层
量少直接上讯飞6麦克阵列,也不贵。量大自己研究划算

出0入0汤圆

发表于 2017-3-24 15:06:35 | 显示全部楼层
高人,膜拜!

出0入0汤圆

 楼主| 发表于 2017-3-24 15:54:02 | 显示全部楼层
smithding 发表于 2017-2-6 14:25
成品,现在优惠价600多。。。。某宝搜索“”ReSpeaker 远场麦克风阵列“”

good                 

出0入0汤圆

发表于 2017-3-24 16:07:53 | 显示全部楼层
不錯的資料,謝謝分享!

出0入0汤圆

发表于 2017-3-24 22:07:49 | 显示全部楼层
研究声阵列拾音

出0入0汤圆

发表于 2017-3-25 10:24:11 | 显示全部楼层
不錯的資料,謝謝分享!

出0入0汤圆

发表于 2017-6-13 16:32:08 | 显示全部楼层
标记一下

出0入22汤圆

发表于 2017-6-13 17:48:49 | 显示全部楼层
国内 在远场识别  科大讯飞 也是菜鸟,,看看 叮咚音响就知道了。 。

感觉7MIC 识别需要的运算 可能真的很大。不说别的 光FFT 就够耗精力了,,还有快速卷积

出0入37汤圆

发表于 2017-6-13 20:42:48 | 显示全部楼层
标记一下,明天慢慢看

出0入0汤圆

发表于 2018-1-5 10:02:45 | 显示全部楼层
好资料,谢谢楼主

出0入0汤圆

发表于 2018-1-30 20:44:17 | 显示全部楼层
路过,标记下

出0入0汤圆

发表于 2018-1-30 21:24:11 | 显示全部楼层
谢谢分享!

出0入0汤圆

发表于 2018-1-30 21:57:20 | 显示全部楼层
好资料,感谢分享,谢谢

出0入0汤圆

发表于 2018-1-31 09:41:20 | 显示全部楼层
本帖最后由 kap 于 2018-1-31 10:11 编辑

声纹识别, 好东西

出0入0汤圆

发表于 2018-1-31 10:22:04 | 显示全部楼层
好资料,谢谢楼主分享。做个记号

出0入0汤圆

发表于 2018-1-31 10:32:22 | 显示全部楼层
这个厉害了,不知道是不是可以做成像美军配的那种玩意 , 当狙击手放了一枪,就会告知士兵是 几点方向打过来的,,,

出0入0汤圆

 楼主| 发表于 2018-1-31 10:40:28 来自手机 | 显示全部楼层
jackiezeng 发表于 2018-1-31 10:32
这个厉害了,不知道是不是可以做成像美军配的那种玩意 , 当狙击手放了一枪,就会告知士兵是 几点方向打过 ...

可以的,之前我说麻烦,是因为了解不深

出0入0汤圆

发表于 2018-1-31 19:47:11 | 显示全部楼层
xivisi 发表于 2018-1-31 10:40
可以的,之前我说麻烦,是因为了解不深

可以讲一讲定位狙击手方位的原理吗

出0入0汤圆

 楼主| 发表于 2018-1-31 20:05:53 | 显示全部楼层
deadline2012 发表于 2018-1-31 19:47
可以讲一讲定位狙击手方位的原理吗

楼主位的资料看懂了,基本原理就清楚了

出0入0汤圆

发表于 2018-1-31 23:35:06 来自手机 | 显示全部楼层
good   不错

出0入0汤圆

发表于 2018-5-6 19:59:36 | 显示全部楼层
感谢分享,好资料~

出0入0汤圆

发表于 2018-5-8 10:39:53 | 显示全部楼层
好资料,谢谢楼主分享

出0入0汤圆

发表于 2018-5-8 11:19:46 | 显示全部楼层
非常好的资料,谢谢
回帖提示: 反政府言论将被立即封锁ID 在按“提交”前,请自问一下:我这样表达会给举报吗,会给自己惹麻烦吗? 另外:尽量不要使用Mark、顶等没有意义的回复。不得大量使用大字体和彩色字。【本论坛不允许直接上传手机拍摄图片,浪费大家下载带宽和论坛服务器空间,请压缩后(图片小于1兆)才上传。压缩方法可以在微信里面发给自己(不要勾选“原图),然后下载,就能得到压缩后的图片】。另外,手机版只能上传图片,要上传附件需要切换到电脑版(不需要使用电脑,手机上切换到电脑版就行,页面底部)。
您需要登录后才可以回帖 登录 | 注册

本版积分规则

手机版|Archiver|amobbs.com 阿莫电子技术论坛 ( 粤ICP备2022115958号, 版权所有:东莞阿莫电子贸易商行 创办于2004年 (公安交互式论坛备案:44190002001997 ) )

GMT+8, 2024-3-28 22:47

© Since 2004 www.amobbs.com, 原www.ourdev.cn, 原www.ouravr.com

快速回复 返回顶部 返回列表