偶然发现seeedstudio更新了他们的英文版说明书，然而中文版还没更新[捂脸]。关于DOA的部分，除了原来的使用ODAS Studio的方法以外，又多加了一些使用Python代码直接进行DOA的章节。

0 安装驱动#

别忘了用source activate doa切到专门的环境里，或者conda create -n doa python=3.6现造一个环境。

安装驱动：

1
proxychains4 git clone https://github.com/respeaker/seeed-voicecard.git
2
cd seeed-voicecard
3
sudo proxychains4 ./install.sh
4
sudo reboot -h now

检查驱动：

1
arecord -L # 输入驱动，即录制
2
null
3
    Discard all samples (playback) or generate zero samples (capture)
4
default
5
ac108
6
dmixer
7
ac101
8
sysdefault:CARD=seeed8micvoicec
9
    seeed-8mic-voicecard,
10
    Default Audio Device
11
dmix:CARD=seeed8micvoicec,DEV=0
12
    seeed-8mic-voicecard,
13
    Direct sample mixing device
14
dsnoop:CARD=seeed8micvoicec,DEV=0
15
    seeed-8mic-voicecard,
16
    Direct sample snooping device
17
hw:CARD=seeed8micvoicec,DEV=0
18
    seeed-8mic-voicecard,
19
    Direct hardware device without any conversions
20
plughw:CARD=seeed8micvoicec,DEV=0
21
    seeed-8mic-voicecard,
22
    Hardware device with all software conversions
23

24
aplay -L # 输出驱动，即播放
25
null
26
    Discard all samples (playback) or generate zero samples (capture)
27
default
28
ac108
29
dmixer
30
ac101
31
sysdefault:CARD=ALSA
32
    bcm2835 ALSA, bcm2835 ALSA
33
    Default Audio Device
34
dmix:CARD=ALSA,DEV=0
35
    bcm2835 ALSA, bcm2835 ALSA
36
    Direct sample mixing device
37
dmix:CARD=ALSA,DEV=1
38
    bcm2835 ALSA, bcm2835 IEC958/HDMI
39
    Direct sample mixing device
40
dsnoop:CARD=ALSA,DEV=0
41
    bcm2835 ALSA, bcm2835 ALSA
42
    Direct sample snooping device
43
dsnoop:CARD=ALSA,DEV=1
44
    bcm2835 ALSA, bcm2835 IEC958/HDMI
45
    Direct sample snooping device
46
hw:CARD=ALSA,DEV=0
47
    bcm2835 ALSA, bcm2835 ALSA
48
    Direct hardware device without any conversions
49
hw:CARD=ALSA,DEV=1
50
    bcm2835 ALSA, bcm2835 IEC958/HDMI
51
    Direct hardware device without any conversions
52
plughw:CARD=ALSA,DEV=0
53
    bcm2835 ALSA, bcm2835 ALSA
54
    Hardware device with all software conversions
55
plughw:CARD=ALSA,DEV=1
56
    bcm2835 ALSA, bcm2835 IEC958/HDMI
57
    Hardware device with all software conversions
58
sysdefault:CARD=seeed8micvoicec
59
    seeed-8mic-voicecard,
60
    Default Audio Device
61
dmix:CARD=seeed8micvoicec,DEV=0
62
    seeed-8mic-voicecard,
63
    Direct sample mixing device
64
dsnoop:CARD=seeed8micvoicec,DEV=0
65
    seeed-8mic-voicecard,
66
    Direct sample snooping device
67
hw:CARD=seeed8micvoicec,DEV=0
68
    seeed-8mic-voicecard,
69
    Direct hardware device without any conversions
70
plughw:CARD=seeed8micvoicec,DEV=0
71
    seeed-8mic-voicecard,
72
    Hardware device with all software conversions

1 测试LED#

说实话，自打买了这个麦阵列，我还从来没试过这12个LED灯[捂脸]，试试吧要不感觉对不起这个板子，另外，运行demo需要gpiozero、spidev、pyusb。

1
proxychains4 git clone --depth 1 https://github.com/respeaker/pixel_ring.git
2
cd pixel_ring
3
proxychains4 pip install -U -e .
4
proxychains4 pip install gpiozero RPi.GPIO
5
python examples/respeaker_4mic_array.py

控制LED的主要代码其实就在函数show()，直接在Python交互环境中输入下面的代码，就可以打开12点方向的灯，并渲染为白色，也就是ARGB(0, 64, 64, 64)。

1
from pixel_ring import pixel_ring
2
from gpiozero import LED
3

4
power = LED(5)
5
power.on()
6

7
pixel_ring.set_brightness(10) # 0-100，其实会被转换为`APA102.dev.global_brightness`的0-31
8
pixel_ring.pattern.show([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 64, 64, 64])

用于控制的数组是一个48位的list，从1点钟方向开始，到12点钟方向结束，每四位控制一个LED灯，非常好操作。

2 装DOA相关库#

添加seeed提供的apt源。（靠谱的公司啊）

1
echo "deb https://seeed-studio.github.io/pi_repo/ stretch main" | sudo tee /etc/apt/sources.list.d/seeed.list
2
proxychains4 curl https://seeed-studio.github.io/pi_repo/public.key | sudo apt-key add -
3
proxychains4 sudo apt update

安装numpy和pyaudio：

1
proxychains4 conda install -y numpy
2
proxychains4 sudo apt-get install -y portaudio19-dev
3
proxychains4 pip install pyaudio

安装编译工具：

1
sudo proxychains4 apt-get install -y swig python-dev libatlas-base-dev build-essential make

安装snowboy（热词检测DOA）：

1
proxychains4 git clone --depth 1 https://github.com/Kitt-AI/snowboy.git
2
cd snowboy
3
python setup.py build
4
python setup.py bdist_wheel
5
pip install dist/snowboy*.whl

安装seeed的voice-engin：

1
proxychains4 git clone https://github.com/voice-engine/voice-engine.git
2
cd voice-engine
3
python setup.py bdist_wheel
4
pip install dist/*.whl

测试ok~：

1
python ~/doa/voice-engine/examples/respeaker_6mic_array_for_pi/kws_doa.py

如果只是想用DOA功能，可以试试seeed的老项目mic_array：

1
proxychains4 git clone https://github.com/respeaker/mic_array.git
2
python ./mic_array/mic_array.py

对比了一下，跟人家introlab/odas项目差的还是比较远的。直观上看就是我们简单的靠fft实现的DOA仅仅能做到平面范围内的方位判断，仅输出一个角度（平面offset），速度慢且不支持声源跟踪等功能。而odas的声源定位项目输出两个值，一个平面offset一个法向offset，还能进行声源分辨，多声源跟踪跟踪。odas项目的理论收录在Lightweight and Optimized Sound Source Localization and Tracking Methods for Opened and Closed Microphone Array Configurations中，值得一读。