Contact:snowboy@kitt.ai
Website:https://snowboy.kitt.ai
Github:https://github.com/kitt-ai/snowboy
Version:1.1.1 (2017-03-24)
注意，2020年8月这个项目已经更新了，在树莓派上配置不那么好用了，仅供参考吧
refer to: http://docs.kitt.ai/snowboy/
New!
Snowboy now offers Hotword as a Service. You can programmatically use our RESTful API Calls to train a hotword model in 3 Easy Steps.
First pick a hotword.
Record it 3 times on your device
Submit the audio files through our RESTful API Calls and a model will be trained and returned.
Upon completion, the device can immediately perform hotword detection.
The following video demonstrates how it’s done using different customized hotwords in both English and Mandarin Chinese.
Introduction
Snowboy is an highly customizable hotword detection engine that is embedded real-time and is always listening (even when off-line) compatible with Raspberry Pi, (Ubuntu) Linux, and Mac OS X.
A hotword (also known as wake word or trigger word) is a keyword or phrase that the computer constantly listens for as a signal to trigger other actions.
Some examples of hotword include “Alexa” on Amazon Echo, “OK Google” on some Android devices and “Hey Siri” on iPhones. These hotwords are used to initiate a full-fledged speech interaction interface. However, hotwords can be used in other ways too like performing simple command & control actions.
In one hacky solution, one can run a full ASR (Automatic Speech Recognition) to perform hotword detection. In this scenario, the device would watch for specific trigger words in the ASR transcriptions. However, ASR consumes a lot of device and bandwidth resources. In addition, it does not protect your privacy when one uses a cloud-based solution. Luckily, Snowboy was created to solve these problems!
Snowboy is:
highly customizable allowing you to freely define your own magic hotword such as (but not limited to) “open sesame”, “garage door open”, or “hello dreamhouse”. If you can think it, you can hotword it!
always listening but protects your privacy because Snowboy does not connect to the Internet or stream your voice anywhere.
light-weight and embedded allowing you to runs it on Raspberry Pi’s consuming less than 10% CPU on the smallest Pi’s (single-core 700M Hz ARMv6).
Apache licensed!
Currently, Snowboy supports:
all versions of Raspberry Pi (with Raspbian based on Debian Jessie 8.0)
64bit Mac OS X
64bit Ubuntu (12.04 and 14.04)
iOS
Android with ARMv7 CPUs
Pine 64 with Debian Jessie 8.5 (3.10.102)
Intel Edison with Ubilinux (Debian Wheezy 7.8)
Tip
For iOS/Android, please check out Snowboy’s GitHub page.
Note
If you get it to work with more devices, OS, or programming languages, feel free to send a pull request to the GitHub repository.
Downloads
You can download pre-packaged Snowboy binaries and their Python wrappers for:
Raspberry Pi with Raspbian 8.0, all versions (1/2/3/Zero)
Pine 64 with Debian Jessie 8.5 (3.10.102) (Pine64)
Intel Edison with Ubilinux (Debian Wheezy 7.8) (Edison)
Or you can check out GitHub to compile a version yourself.
Note
RPi3 has an ARMv8 CPU but Raspbian recognizes it as ARMv7: The Cortex-A53 can run ARMv7 code just fine (in fact, substantially better than Pi 2 due to architectural improvements). This is where we are staying in terms of supported userspace and kernel for the time being. If you want to roll your own OS/kernel, you can add arm_control=0x200 to /boot/config.txt to boot the cores in ARMv8 state.
Quick Start
To use Snowboy, you’ll need:
A supported device with a microphone (or a microphone input)
The corresponding decoder (downloaded above)
A trained model(s) from https://snowboy.kitt.ai.
Access Microphone
We use PortAudio as a cross-platform support for audio in/out. We also use sox as a quick utility to check whether the microphone setup is correctly.
Install Sox.
On Linux systems, run:
sudoapt-getinstallpython-pyaudiopython3-pyaudiosox
On Mac, run:
brewinstallportaudiosox
Note
If you don’t have Homebrew, install it here
Install PortAudio’s Python bindings:
pipinstallpyaudio
Note
If you don’t have pip, you can install it here
Tip
If you have a Permission Error from pip, you can either use sudo pip install pyaudio or change the folder owner to yourself: sudo chown $USER -R /usr/local
To check whether you can record via your microphone, open a terminal and run:
rectemp.wav
Tip
If you see an error on Raspberry Pi’s, please refer to the Running_on_Pi section.
Decoder Structures
The decoder tarball contains the following files:
├── README.md├── _snowboydetect.so├── demo.py├── demo2.py├── light.py├── requirements.txt├── resources│ ├── ding.wav│ ├── dong.wav│ ├── common.res│ └── snowboy.umdl├── snowboydecoder.py├── snowboydetect.py└── version
_snowboydetect.so is a dynamically linked library compiled with SWIG. It has dependencies on your system’s Python2 library. All snowboy-related libraries are statically linked in this file.
snowboydetect.py is a Python wrapper file generated by SWIG. Because it is not very easy to read, we created the other high-level wrapper: snowboydecoder.py.
You should already have a trained model file from https://snowboy.kitt.ai (for example snowboy.pmdl), or you can simply use the universal model in resources/snowboy.umdl.
Running a Demo
Tip
This demo runs on any devices. But we suggest you run it on a laptop/desktop with speaker output because the demo plays a Ding sound when your hotword is triggered.
To access the simple demo in __main__ code of snowboydecoder.py, run the following command in your Terminal:
pythondemo.pysnowboy.pmdl
Here snowboy.pmdl is your trained model downloaded from https://snowboy.kitt.ai.
Note
The .pmdl suffix indicates a personal model and a .umdl suffix indicates a universal model.
When prompt, speak into your microphone to see whether snowboy detects your magic phrase.
The demo is fairly straight-forward. The following is the demo’s code:
importsnowboydecoderimportsysimportsignalinterrupted=Falsedefsignal_handler(signal,frame):globalinterruptedinterrupted=Truedefinterrupt_callback():globalinterruptedreturninterruptediflen(sys.argv)==1:print("Error: need to specify model name")print("Usage: python demo.py your.model")sys.exit(-1)model=sys.argv[1]signal.signal(signal.SIGINT,signal_handler)detector=snowboydecoder.HotwordDetector(model,sensitivity=0.5)print('Listening... Press Ctrl+C to exit')detector.start(detected_callback=snowboydecoder.ding_callback,interrupt_check=interrupt_callback,sleep_time=0.03)detector.terminate()
The main program loops at detector.start(). Every sleep_time=0.03 second, the function:
checks a ring buffer filled with microphone data to see whether a hotword is detected. If yes, call the detected_callback function.
calls the interrupt_check function: if it returns True, then break the main loop and return.
Here, we assigned detected_callback with a default snowboydecoder.ding_callback so that every time your hotword is heard the computer will play a ding sound.
Warning
Do not append () to your callback function: the correct way is to assign detected_callback=your_func instead of detected_callback=your_func(). However, what if you have parameters to assign in your callback functions? Use a lambda function! So your callback would look like: callback=lambda: callback_function(parameters).
Running on Raspberry Pi
Raspberry Pi’s are excellent hardware for running Snowboy. We support all versions of Raspberry Pi (1, 2, 3 and Zero). Supported OS is Raspbian 8.0.
Set up Audio
Warning
You’ll need a USB microphone for audio input. The on-board 3.5mm audio jack only has audio out but no audio in thus a microphone with a 3.5mm audio jack will not work.
Tip
We have successfully used both generic USB microphones and the PlayStation 3 Eye webcam. You can buy a PS 3 Eye for $5 on Amazon. Linux has builtin kernel modules for it but Windows PCs do not have free drivers for the Eye.
Before beginning, please follow Access_Microphone to install portaudio to test whether your microphone can be accessed with rec:
rect.wav
Warning
even though USB webcams should be “plug-n-play”, we experienced that for some of them you have to reboot the Pi after plugging in the webcam.
If you see errors, check whether your alsa/pulseaudio is configured properly. First list the playback device:
$ aplay -l **** List of PLAYBACK Hardware Devices **** card 0: ALSA [bcm2835 ALSA], device 0: bcm2835 ALSA [bcm2835 ALSA] Subdevices: 8/8 Subdevice #0: subdevice #0 Subdevice #1: subdevice #1 Subdevice #2: subdevice #2 Subdevice #3: subdevice #3 Subdevice #4: subdevice #4 Subdevice #5: subdevice #5 Subdevice #6: subdevice #6 Subdevice #7: subdevice #7 card 0: ALSA [bcm2835 ALSA], device 1: bcm2835 ALSA [bcm2835 IEC958/HDMI] Subdevices: 1/1 Subdevice #0: subdevice #0
Here the playback device is card 0, device 0, or hw:0,0 (hw:0,1 is HDMI audio out).
List your recording device:
$ arecord -l**** List of CAPTURE Hardware Devices ****card 1: Camera [Vimicro USB2.0 UVC Camera], device 0: USB Audio [USB Audio] Subdevices: 1/1 Subdevice #0: subdevice #0
Here the recording device is card 1, device 0, or hw1:0.
Change your ~/.asoundrc file to:
pcm.!default { type asym playback.pcm { type plug slave.pcm "hw:0,0" } capture.pcm { type plug slave.pcm "hw:1,0" }}
Try rec temp.wav again. Your microphone in should be set up properly now.
Go back to Running a Demo and run the demo.
If the demo runs is successful, try to Blink an LED light and Toggle an AC-powered Lamp with your Pi.
Warning
If you see the following error:
ImportError: /usr/lib/arm-linux-gnueabihf/libstdc++.so.6:version `GLIBCXX_3.4.20' not found (required by rpi-arm-raspbian-8.0-1.0.1/_snowboy.so)`
It means that your g++ library is not up-to-date. You are probably still using Debian Wheezy 7.5 (check with lsb_release -a). However we compiled the snowboy library under Raspbian based on Debian Jessie 8.0 that comes with g++-4.9. You can either upgrade your Raspbian version to Jessie, or follow this post to install g++-4.9 on your Wheezy, or compile a version youself from GitHub.
Tip
If you cannot hear any audio from the 3.5mm audio jack, the audio may be streamed to the HDMI port. Follow this config to change the audio output to the 3.5mm audio jack.
Blink an LED light
Wire an LED
Wiring an LED onto Pi’s GPIO ports is very easy. However, note that the LED has a shorter leg and a longer leg. The shorter leg is usually connected to ground (GND). The following demonstrates how to wire your LED to your Pi’s GPIO:
A few hundred Ohms would be enough for the resistor.
Control an LED with Python
We use the RPi.GPIO Python module to control and LED:
importRPi.GPIOasGPIOimporttimeclassLight(object):def__init__(self,port):self.port=portGPIO.setmode(GPIO.BCM)GPIO.setup(self.port,GPIO.OUT)self.on_state=GPIO.HIGHself.off_state=notself.on_statedefset_on(self):GPIO.output(self.port,self.on_state)defset_off(self):GPIO.output(self.port,self.off_state)defis_on(self):returnGPIO.input(self.port)==self.on_statedefis_off(self):returnGPIO.input(self.port)==self.off_statedeftoggle(self):ifself.is_on():self.set_off()else:self.set_on()defblink(self,t=0.3):self.set_off()self.set_on()time.sleep(t)self.set_off()if__name__=="__main__":light=Light(17)whileTrue:light.blink()time.sleep(0.7)
Save the file as light.py, then run:
sudopythonlight.py
The LED light will blink approximately once per second.
Blink an LED with Snowboy
Replace Snowboy’s callback function with LED’s blink() function:
importsnowboydecoderimportsysimportsignalfromlightimportLightinterrupted=Falsedefsignal_handler(signal,frame):globalinterruptedinterrupted=Truedefinterrupt_callback():globalinterruptedreturninterruptediflen(sys.argv)==1:print("Error: need to specify model name")print("Usage: python demo.py your.model")sys.exit(-1)model=sys.argv[1]signal.signal(signal.SIGINT,signal_handler)detector=snowboydecoder.HotwordDetector(model,sensitivity=0.5)print('Listening... Press Ctrl+C to exit')led=Light(17)detector.start(detected_callback=led.blink,interrupt_check=interrupt_callback,sleep_time=0.03)detector.terminate()
The only place that changes is:
led=Light(17)detector.start(detected_callback=led.blink,interrupt_check=interrupt_callback,sleep_time=0.03)
which will blink the LED connected to GPIO pin 17 when your hotword is detected:
sudopythondemo.pyyour.pmdl
Toggle an AC-powered Lamp
Controlling an LED light is pretty simple so let’s go bigger and control some real home appliances!
In this example, we will use Raspberry Pi’s GPIO output to connect and break a higher voltage AC circuit.
This can be done with help of a bipolar transistor. Luckily, one has already been built thanks to a successful kickstarter campaign. You can purchase the IoT relay on Amazon for $15 (as of April 2016).
The mechanism of the IoT Relay is very simple:
When red wire has high DC voltage (say, 3.3V or 12V), the top two “normally ON” outlets will turn off and the bottom two “normally OFF” outlets will turn on
When red wire has no DC voltage, the top two “normally ON” outlets will turn on and the bottom two “normally OFF” outlets will turn off
Note
The top two and bottom two outlets can only be controlled in two groups. There is no way to control each of them individually.
To connect the IoT Relay to your Raspberry Pi, connect the red wire of the IoT Relay to Pin 17 of a Raspberry Pi. You can simply reuse light.py or demo.py above to control any home appliances that are plugged into the IoT Relay!
The following is demonstrates Snowboy on a Raspberry Pi controlling three small LED lights on the right and a lamp on the left through the IoT relay:
RESTful API Calls
Snowboy provides the following HTTP endpoints for you to train a model without using the website:
/api/v1/train: train a model with 3 .wav files
All uploaded .wav files will not be visible on the Snowboy library, so your privacy is protected. However, we do not provide an API to retrieve these files either.
/api/v1/train
The /api/v1/train endpoint provides an opportunity to:
programmatically train a model without using the web interface
achieve better acoustic consistency
Note
Since the training and test voice samples will be collected off the same microphone, there will be no distortions that result from the usage of different microphones
You can define truly customized hotword for each of your end customer. Just ask them to say the hotword 3 times and a model will be trained on the fly!
Endpoint: https://snowboy.kitt.ai/api/v1/train/
Type: POST
Return: a binary personal model (.pmdl), or error
ParameterRequiredValue
voice_samplesYA list of 3 voice samples in .wav format.
tokenYSecret user token
nameYString, or “unknown” if we don’t know hotword name
languageNar (Arabic), zh (Chinese), nl (Dutch), en (English), fr (French), dt (German), hi (Hindi), it (Italian), jp (Japanese), ko (Korean), fa (Persian), pl (Polish), pt (Portuguese), ru (Russian), es (Spanish), ot (Other)
age_groupN0_9, 10_19, 20_29, 30_39, 40_49, 50_59, 60+
genderNF/M
microphoneNString, your microphone type
Note
API token can be obtained by logging into https://snowboy.kitt.ai, click on “Profile settings”:
The following is a sample call script using Python. Save the file as training_service.py:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
importsysimportbase64importrequestsdefget_wave(fname):withopen(fname)asinfile:returnbase64.b64encode(infile.read())endpoint="https://snowboy.kitt.ai/api/v1/train/"############# MODIFY THE FOLLOWING #############token=""hotword_name="???"language="en"age_group="20_29"gender="M"microphone="macbook microphone"############### END OF MODIFY ##################if__name__=="__main__":try:[_,wav1,wav2,wav3,out]=sys.argvexceptValueError:print"Usage: %s wave_file1 wave_file2 wave_file3 out_model_name"%sys.argv[0]sys.exit()data={"name":hotword_name,"language":language,"age_group":age_group,"gender":gender,"microphone":microphone,"token":token,"voice_samples":[{"wave":get_wave(wav1)},{"wave":get_wave(wav2)},{"wave":get_wave(wav3)}]}response=requests.post(endpoint,json=data)ifresponse.ok:withopen(out,"w")asoutfile:outfile.write(response.content)print"Saved model to '%s'."%outelse:print"Request failed."printresponse.text
To execute, run the following command:
pythontraining_service.py1.wav2.wav3.wavsaved_model.pmdl
..note:: You can use the rec command to record a .wav file on terminal:
rec-r16000-c1-b16-esigned-integer1.wav
The following is a sample call script in bash with curl:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
#! /usr/bin/env bashENDPOINT="https://snowboy.kitt.ai/api/v1/train/"############# MODIFY THE FOLLOWING #############TOKEN="??"NAME="??"LANGUAGE="en"AGE_GROUP="20_29"GENDER="M"MICROPHONE="PS3 Eye"############### END OF MODIFY ##################if[["$#"!=4]];thenprintf"Usage: %s wave_file1 wave_file2 wave_file3 out_model_name"$0exitfiWAV1=`base64$1`WAV2=`base64$2`WAV3=`base64$3`OUTFILE="$4"cat<<EOF >data.json { "name": "$NAME", "language": "$LANGUAGE", "age_group": "$AGE_GROUP", "token": "$TOKEN", "gender": "$GENDER", "microphone": "$MICROPHONE", "voice_samples": [ {"wave": "$WAV1"}, {"wave": "$WAV2"}, {"wave": "$WAV3"} ] } EOFcurl -H"Content-Type: application/json"-X POST -d @data.json$ENDPOINT>$OUTFILE
Quota and Rate Limit
Each user gets 1000 free API calls for each end point every 30 days with a rate limited of 1 call per second.
If you’d like to purchase more, please send us an email at snowboy@kitt.ai
Advanced Usage
Multiple Models and Callbacks
So far we have worked with only one model that can dictate a binary state. Wouldn’t it nice to listen to multiple models at the same time?
demo2.py demonstrates how to listen to multiple models at the same time:
importsnowboydecoderimportsysimportsignalinterrupted=Falsedefsignal_handler(signal,frame):globalinterruptedinterrupted=Truedefinterrupt_callback():globalinterruptedreturninterruptediflen(sys.argv)!=3:print("Error: need to specify 2 model names")print("Usage: python demo.py 1st.model 2nd.model")sys.exit(-1)models=sys.argv[1:]# capture SIGINT signal, e.g., Ctrl+Csignal.signal(signal.SIGINT,signal_handler)sensitivity=[0.5]*len(models)detector=snowboydecoder.HotwordDetector(models,sensitivity=sensitivity)callbacks=[lambda:snowboydecoder.play_audio_file(snowboydecoder.DETECT_DING),lambda:snowboydecoder.play_audio_file(snowboydecoder.DETECT_DONG)]print('Listening... Press Ctrl+C to exit')# main loop# make sure you have the same numbers of callbacks and modelsdetector.start(detected_callback=callbacks,interrupt_check=interrupt_callback,sleep_time=0.03)detector.terminate()
In this example, we used two models for the decoder and provided two callback functions. If the first hotword is detected, it’ll play a Ding sound. If the second hotword is detected, it’ll play a Dong sound.
Note
You are not limited to just using only two models nor are you limited to using only the personal or the universal models. You can give HotwordDetector a mixture of multiple personal and universal models so long as your CPU is powerful enough to process them all.
FAQ
What’s the CPU/RAM usage?
Snowboy takes minimal CPU on modern computers. On a Raspberry Pi’s with decade-old CPU chips, it takes less than 5% ~ 10% of CPU. In terms of memory usage, the PortAudio Python wrapper usually uses about 10MB of RAM while the standalone C binary uses less than 2MB.
NameCPUCPU UsageRAM Usage
RPi 1single-core 700MHz ARMv6<10%Python: < 15MB
C: < 2MB
RPi 2quad-core 900MHz ARMv7<5%
RPi 3quad-core 1.2GHz ARMv8<5%
RPi Zerosingle-core 1GHz ARMv6<5%
MacbooksIntel Core i3/5/7<1%
What is detection sensitivity
Detection sensitivity controls how sensitive the detection is. It is a value between 0 and 1. Increasing the sensitivity value lead to better detection rate, but also higher false alarm rate. It is an important parameter that you should play with in your actual application.
What Audio format does Snowboy support?
Snowboy supports WAVE files (with linear PCM, 8-bits unsigned integer, 16-bits signed integer or 32-bits signed integer). See SampleRate(), NumChannels() and BitsPerSample() for the required sampling rate, number of channels and bits per sample values.
To convert your .wav file to Snowboy supported format, you can use sox:
sox -t wav YOUR_ORIGINAL.wav -t wav -r 16000 -b 16 -e signed-integer -c 1 YOUR_PROCESSED.wav
My pmdl model works well for me, but does not work well for others
Models with suffix pmdl are personal models thus they are supposed to only work well for the person who provides the audio samples. If you are looking for a model that works well for everyone, you should use the universal model (with suffix umdl).
My trained model works well on laptops but not on Pi’s
This is due to the acoustic distortion that results from the different microphones. If you record your voice with two different microphones (one on your laptop and the other on your Pi) and then play them (play t.wav), you will hear that they sound very differently (even though it is the same voice)!
The best solution is to use the same recording to both train your model and test your voice. If you want to use Snowboy on a Raspberry Pi, first record your voice with rec t.wav (make sure to apt-get install sox) and then upload the 3 recordings to the Snowboy website using uploading button:
Once the training has completed, you can download the trained model.
Alternatively, you can also use the RESTful API Calls to do this directly without using the web interface.
Tip
Another trick is to play with the audio gain (see the answer regarding audio_gain below). We have noted that the USB microphones on a Raspberry Pi usually have low volume, thus increasing the audio gain may help.
The volume of my recording is too low/high
When you construct a HotwordDetector from snowboydecoder, there is an audio_gain parameter:
HotwordDetector(decoder_model,resource=RESOURCE_FILE,sensitivity=[],audio_gain=1)
Set audio_gain to be larger than 1 if your test recording’s volume is too low, or smaller than 1 if too high.
Does Snowboy come with VAD?
Yes it does! VAD is Voice Activity Detection which usually detects whether there’s human voice in the audio. It needs much less resources than hotword detection. Thus, Snowboy uses VAD as a filtering layer before hotword detection to reduce CPU usage.
How to use Snowboy’s VAD to detect voice and silence?
The return value of SnowboyDetect.RunDetection() function indicates silence, voice, error, and triggered words:
returnmeaning
-2silence
-1error
0voice
1,..triggered index
Check out snowboydecoder.py for usages.
Who wrote Snowboy?
The KITT.AI co-founders. Core modules of Snowboy are created by Guoguo Chen, who is also a contributor to the open-source speech recognition software Kaldi and Microsoft Cognitive Toolkit CNTK.
网友评论