PRAAT语音分析软件使用教程(英文版)Using Praat for Linguistic Research Latest
Using Praat for Linguistic Research
Will Styler
University of Colorado at Boulder Phonetics Lab
Document Version:1.3.6
Last Update:October2,2013
Important!
This document will be continually updated and improved.Download the latest version at:
https://www.sodocs.net/doc/e517798544.html,/praat
Using Praat for Linguistic Research by Will Styler is licensed under a Creative Commons Attribution-ShareAlike3.0Unported License.For more information on the speci?cs of this license,go to https://www.sodocs.net/doc/e517798544.html,/licenses/by-sa/3.0/.
1
Contents
1Version History4 2Introduction5
2.1Versions (5)
2.2Other Resources (5)
2.3Getting and Installing Praat (6)
3About Praat6
3.1Praat Windows (6)
4Recording Sounds8 5Opening and Saving Files8
5.1Opening Files (8)
5.1.1Working with longer sound?les (9)
5.2Saving Files (9)
6Phonetic Measurement and Analysis in Praat9
6.1Working with Praat Waveforms and Spectrograms (9)
6.1.1Pulling out a smaller section of the?le for analysis (11)
6.2Adjusting the Spectrogram settings (11)
6.2.1Narrowband vs.Broadband Spectrograms (12)
6.3Measuring Duration (13)
6.3.1Measuring Voice Onset Time(VOT) (13)
6.4Examining and measuring F0/Pitch (14)
6.4.1Measuring F0from a single cycle (14)
6.4.2Viewing Pitch via a narrowband spectrogram (14)
6.4.3Using Praat’s Pitch Tracking (15)
6.4.4Improving Pitch tracking by changing the Pitch Settings (15)
6.4.5Scripting only:Creating a Pitch Object (17)
6.4.6Getting Maximum,Minimum,and Average pitch for a section of speech.17
6.5Measuring Pulses,Jitter,Shimmer,and Harmonics-to-noise ratio (17)
6.6Measuring Formants (18)
6.6.1Using the Formant tools in the Editor window (18)
6.6.2Improving Formant Finding results (19)
6.6.3Scripting Only:Formant Objects (21)
6.7Measuring Intensity/Amplitude (21)
6.8Working with Spectra (22)
6.9Taking a spectral slice (22)
6.10Measuring Harmonic Amplitude,Frequency (23)
2
6.11Measuring Creakiness and Breathiness using Spectral Tilt (23)
6.12Measuring Nasality using A1-P0 (24)
6.13Measuring Spectral Center of Gravity (25)
7Creating and manipulating sound Files in Praat27
7.1Creating sounds from Formula (27)
7.2Cropping,Copying,Splicing and Pasting (28)
7.3Sampling rates and Resampling (29)
7.4Filtering Sounds (30)
7.4.1Low-pass?ltering (30)
7.4.2High-pass?ltering (31)
7.4.3Band-pass(notch)?ltering (31)
7.5Pitch Manipulation(To Manipulation...). (32)
7.6Matching the pitch tracks of two sounds (32)
7.7Manipulating Duration(Slowing Down and Speeding Up Sounds) (34)
7.8Matching Intensity (35)
7.9Combining Sounds (35)
7.10Formula Modi?cation:Waveform addition,subtraction and so much more (36)
7.11Synthesizing Sounds from scratch (39)
7.12Source-Filter Vowel Resynthesis (39)
8Exporting images for use and publication41
8.1Creating Complex Displays (43)
8.1.1Overlaying Plots (43)
8.1.2Multiple Plots in the Picture Window (44)
9Annotating Sound Files(Praat TextGrids)44 10Using Log Files49 11Scripting in Praat50
11.1What is Praat scripting? (50)
11.1.1Praat’s scripting tutorials (52)
11.1.2Praat scripts vs.Editor scripts (52)
11.2Working with Scripts (53)
11.2.1Opening and running a Praat script (53)
11.2.2Making(and removing)Menu Shortcuts for scripts (54)
11.3Creating a new script (56)
11.3.1Using other text editors (56)
11.3.2Filenames (57)
11.3.3How to magically write a Praat script(using the Praat"history"function)57
11.3.4Writing your?rst single-command script (59)
3
11.3.5Scripts with Variables (60)
11.4About the Praat Scripting Language (62)
11.4.1’for’loops (62)
11.4.2’if’statements (63)
11.4.3Forms (64)
11.4.4Commented lines(#) (66)
11.4.5Useful tips (67)
11.4.6Everything Else (68)
11.5In defense of Code Cannibalism (68)
11.6Closing Remarks on Praat scripting (69)
1Version History
?1.3.6-October2,2013-Added discussion of the smoothing of the intensity line and its relation to the pitch track in Praat.
?1.3.5-March22,2013-Updated License for the Manual.A few typoes squashed.
?1.3-September28th,2012-Added Section7.7,Manipulating Duration with explana-tions of how to slow and speed sound?les
?1.2.5-August10th,2012-Added discussion of Praat Picture for more complex displays ?1.2-May19th,2012-A few small tweaks,added a section on measuring Voice Onset Time
?1.1.1-January2nd,2012-A few other small corrections based on Paul Boersma’s feedback.
?1.1-January1st,2012-Updated with Paul Boersma’s valuable feedback and a variety of small corrections.
?1.0.2-July14th,2011-Removed some typoes and?xed other small issues.
?1.0.1-July10th,2011-Updated to include instructions on removing scripts from But-tons.
?1.0-July10th,2011-Version created for LSA Institute Workshop on Praat.
4
2Introduction
Praat is a wonderful software package written and maintained by Paul Boersma and David Weenink of the University of Amsterdam.Available for free,with open source code,there is simply no better package for linguists to use in analyzing speech.
Unfortunately,much of the existing documentation for the software is just that,software docu-mentation,and is not designed to help linguists(who may not necessarily consider themselves to be"phoneticians"or have a strong phonetics background)get the measurements and make the changes that they need and desire for their research.
As such,rather than introducing each menu item and function as such,I’ve instead chosen to describe how to do some of the tasks that linguists want to do without assuming a strong phonetics or programming background.Then,eventually,we’ll discuss some of the more com-plicated measures and tricks one can perform with Praat.
Of course,no one workshop can discuss the myriad of features present in Praat,nor cover all of the quirks of the package,but this workshop will hopefully leave you feeling more at home in Praat,and give you an opportunity to go forth and explore further on your own.
2.1Versions
This guide will assume that you are using a relatively recent build of Praat(5.2+),and there’s no reason not to download the latest version whenever released.All screenshots will be from Praat5.2.27running on Mac OS X,but your copy on your platform should not differ signi?-cantly.Unless otherwise speci?ed,work?ows for making measurements and manipulations do not differ signi?cantly across platforms.
2.2Other Resources
Although this guide aims to be painfully comprehensive,there are many other resources avail-able for helping with Praat.The?rst step for dealing with any issue is Praat’s built in help guide,accessible from the upper right corner of most windows in the program.You’ll be best served by starting with"Intro"and moving from there.
There are also a variety of tutorials for Praat available online,and the Yahoo!Groups"Praat-users"group,whose archives can be searched at the below link:
https://www.sodocs.net/doc/e517798544.html,/group/praat-users/
5
2.3Getting and Installing Praat
Praat can be downloaded from https://www.sodocs.net/doc/e517798544.html,,and its installation will vary depend-ing on your platform.
?Mac OS X-Just drag Praat into your Applications folder.
?Windows(XP,Vista,7)-Download the installer and run it,and a link to the program will be placed on your desktop.
?Linux-If you’re running Linux,you’ll be able to?gure out the install on your own.
Many distributions have Praat as an installable package in their repositories,but check the version numbers,as you won’t want anything older than5.2.x.
3About Praat
3.1Praat Windows
Once you’ve opened Praat,a variety of windows will open automatically,and there are many other windows which will pop up when using the software.It’s best to discuss these now so we can refer to them by name later when discussing the path to certain commands.
The Praat Objects window(Figure1)is where you’ll start most work?ows,using this menu to open,create and save?les,as well as to open the various editors and queries which you’ll need to work with sound?les.
The Editor window(Figure2)is where you’ll spend most of your time,and can be accessed by selecting a sound and choosing"View&Edit".When examining a sound?le,the editor window will show the sound’s waveform on the top and a spectrogram on the bottom,and the cursor will allow you to take selections and measurements.The menus along the top will allow you to show and hide different bits of information(formants,pitch,intensity),as well as to make more detailed queries.When working with other types of Praat objects(e.g.spectra), the editor window will allow you to query those objects as well.
When you make a query,either in the editor window or from the objects window,the Info Window will pop up with your results.You can also print to this window when scripting in Praat(see Section11).Note that information printed here will not necessarily be saved,and running a new query will overwrite it by default.
The Praat Picture window(shown towards the end of the document in Figure10)is used to create and display publication-quality images,and is open by default when you start the program.For detailed information about using the Pictures window and why it exists at all, see Section8.
6
7
Knowing the names of all these commands allow us to more easily describe the commands to use when working with Praat.For instance,if this guide says that you get the duration of a sound by using Objects→Query→Query Time Domain→Get Total Duration,that means, roughly,"Go to the Objects window,Choose"Query",then from that submenu choose"Query Time Domain",then"Get Total Duration"".
4Recording Sounds
To record sound using Praat,you’ll want to plug in your microphone,sound card,or external ADC(Analog-Digital Conversion)box to your computer before starting Praat,and then... Objects→New→Record Mono
This will pull up a recording menu which allows you to choose a sampling frequency(the de-fault,44100Hz,is?ne for most purposes),a microphone or other sound source,and whether to record a mono or stereo sound.Press Record to record,and Stop to stop,being careful that the sound level bar stays within the"green"range to avoid clipping.Once you’ve made a recording,name it and choose Save to list,and it will now show up in the Praat objects window where it’s ready for editing.
If you don’t see a green bar(indicating that Praat hears you)while you are recording,try changing the Input source on the left side of the SoundRecorder window.If this doesn’t help, go to the computer’s sound control panel to ensure the proper microphone is selected,and that the input volume is not turned way down.
Praat only records in one minute long chunks out of the box.To record longer sounds,you can either change the buffer size in Praat→Preferences→Sound Recording Preferences or you can use Audacity(free,available from https://www.sodocs.net/doc/e517798544.html,/)to record the session and then import the sounds into Praat(see Section5)afterwards for analysis and manipulation.
5Opening and Saving Files
5.1Opening Files
If you already have a sound?le recorded that you’d like to open(recorded in.aiff,.wav or.?ac format),there are two ways to open it in Praat.If you’re using Praat on OS X,you can drag supported?les onto the Praat icon in the dock.However,if that doesn’t work,or if you’re on a different platform:
8
Objects→Open→Read from File...
Then use the next dialog to?nd the?les you’re interested in on your hard disk.Once you’ve loaded the?les,they’ll appear in your objects window for further use.Note that other?les created by Praat can be opened in the same way.
Tip:Praat can’t open.wma,.mp3or.m4a audio?les.To convert these easily to.wav?les en masse,download iTunes,set it to import?les in.wav or.aiff format in Preferences,and use iTunes →Advanced→Create.wav version.
5.1.1Working with longer sound?les
Praat has historically had trouble working on sound?les more than20minutes long,and if you’re using a32-bit version or have little available memory,you may experience frequent out-of-memory errors working with large?les unless you use the Objects→Open→Open long sound?le...option.However,sounds read in as Long Sound objects cannot be used for all measurements.It’s recommended to cut your?les into chunks shorter than an hour,either using Audacity or by editing the Long Sound object as described in Section7.2.
5.2Saving Files
Praat does not save anything by default,and until you save?les explicitly,opened or edited ver-sions of the?les will exist?eetingly in the objects window.For emphasis,unless you save?les explicitly,they will disappear completely and unrecoverably when Praat is closed.
To save a?le,select the?le in the Objects window,then...
Objects→Save→Save as_____?le
For sound?les,you’ll likely choose Save as WAV?le,but for the other types of?les(textgrids, formant objects,pitch objects,etc),you’ll save them as text?les.
As this can be very tedious,you might consider downloading and installing a Praat script(see Section11)which saves all the objects in the objects window at once.
6Phonetic Measurement and Analysis in Praat
6.1Working with Praat Waveforms and Spectrograms
Once a sound has been recorded or opened,you’ll spend much of your time interacting with the sound by means of the Editor window.To open a sound in the editor window,select the
9
Figure2:The Praat Editor Window
sound and then...
Objects→View&Edit
You’re immediately presented with an editor window(like that in Figure2),showing the wave-form of the sound,and if the sound is suf?ciently short,a broadband spectrogram showing the spectral energy of the sound over time.In addition,you might also be presented with a series of red dots(representing formants),blue lines(representing the speaker’s pitch),and a yellow line(representing intensity).These can be enabled and disabled in the Editor→View→Show Analyses menu.
Clicking within this window will place the cursor on the waveform and spectrogram.If you click within the editor window,the cursor will spawn two dotted lines.A vertical bar shows the time within the sound where you clicked(labeled at the top in seconds)and,if you clicked within the spectrogram,a horizontal bar shows the frequency at the cursor(labeled on the left in red).If the pitch or intensity tracks are displayed where the cursor is placed,values at the time the cursor represents are given on the left side of the editor window.
In addition,you can click and drag(or use the Select menu)to select portions of the sound. The time of the start and?nish of the selection will be displayed in red,and the duration of the selection(in seconds)will be displayed in the top of the bar.
To play a sound in the editor window,use the three gray bars at the bottom of the editor window.The bottom-most bar(Total Duration)will play the entire sound.The middle bar
10
(Visible Part)will play only the visible portion of the sound.The different sections of the top bar(split by the cursor or selection),when clicked,will play the corresponding pieces of the visible portions of the sound?le.Hitting
Obviously,to view some analyses and to get a closer look at your data,you’ll need to use the ?ve buttons in the bottom left corner of the window.As you can imagine,all shows the entire ?le,in and out zoom in and out,sel zooms to make the current selection?ll the window,and bak zooms back to the previous zoom level.For longer sound?les,in order to view analyses like the spectrogram and formants,you’ll need to zoom in to show only a pre-de?ned amount of time.1
The Group setting in the bottom right corner of the window will ensure that if two sounds are open in Editor windows at once,they’ll share the same zoom characteristics.This is best used to compare two versions of the same?le,say,an original versus one with an acoustic modi?cation made.
All of the measures discussed in this section will use the Editor window,and you will spend much of your time working with Praat here,so any time spent gaining familiarity will be repaid in spades.
6.1.1Pulling out a smaller section of the?le for analysis
Although zooming in and out will get you most of the way there,it’s often useful to isolate a section of a sound(usually a single word or vowel)into a different Sound object.To do this, select a portion of a sound,say,a vowel,and then:
Editor→File→Extract Selected Sound(time from0)
This will create a new sound in the Objects window,containing just the selected part of the original sound.The(preserve times)option(in the same Editor→File menu)just keeps the timecode on the extracted sound the same as in context(so,if the vowel starts at0.245s,the extracted sound?le will start at0.245s).
This can also be done from the objects window using Objects→Convert→Extract part...
6.2Adjusting the Spectrogram settings
Although the basic0-5000Hz broadband spectrogram will suf?ce for many uses,it’s useful to be able to change those settings.To make changes to the spectrogram settings...
1This amount of time can be changed in Editor→View→Show Analyses→Longest analysis.20seconds is a sane value for most modern computers,much higher will cause your system to lag when viewing?les.
11
Figure3:Praat’s Spectrogram Settings
Editor→Spectrum→Spectrogram Settings
This will pull up the Spectrogram settings window(like that in Figure3)
The most important settings here are the window length and view range.
View range controls how much of the spectrum is visible.For speech,you’ll likely be interested in the range from0to5000or6000Hz,but if you’re examining fricatives,you might want to look as high as15,000Hz.If you’re looking at music,you may focus on the area from100to 2000Hz.Either way,this is how you set which part of the spectrum you care about.
If your sound?les have a relatively small or large dynamic range(the difference in volume between the loudest and quietest parts),or if your spectrograms seems too light or too dark, you may want to adjust the dynamic range setting here,but50dB is usually safe.
Window length(given in seconds)controls how large of a chunk of the sound Praat will ex-amine when trying to?nd the frequencies present at a given moment in the signal.Looking at a larger window of the sound will give more accurate information about the frequencies present,but will also reduce the accuracy of the temporal information given.Varying the win-dow length allows you to choose between Broadband and Narrowband spectrograms.
6.2.1Narrowband vs.Broadband Spectrograms
Praat defaults to showing a Broadband spectrogram,which is excellent for viewing the tempo-ral structure of the sound and for seeing vowel formants,but sometimes,you’ll want to look at harmonics and F0instead.To do this,you’ll ask Praat to provide you with a narrowband spectrogram.To do this:
1.Editor→Spectrum→Spectrogram Settings
12
2.Set the Window Length to0.025(or the narrowband window length of your choosing)
3.Click OK
Now,harmonics should clearly be visible in the spectrogram.To return to a broadband spec-trogram:
1.Editor→Spectrum→Spectrogram Settings
2.Set the Window Length to0.005(or the broadband window length of your choosing)
3.Click OK
And you’ll be back to the default broadband spectrogram.
6.3Measuring Duration
As you might expect,measuring duration is quite easy.Once the sound?le is open in the Editor window:
1.Select the portion of the?le you’d like to measure(e.g.the vowel)with the cursor
2.Read the duration of the selection(in seconds)from the duration bar along the bottom
of the Editor window OR
3.Editor→Query→Get selection length and read your selection in the info window
If you’d like the duration of an entire?le,just select the?le in the Objects window and: Objects→Query→Query Time Domain→Get Total Duration
6.3.1Measuring Voice Onset Time(VOT)
"Voice Onset Time"(VOT)is the time between when the stop is released and when the voicing of the following vowel begins.Measuring this time,which can be positive(say,for the English voiceless aspirated stop[t?a]),around zero(for the English"voiced"stop/d/,or,more com-monly,the voiceless unaspirated[ta]around the the world),or negative(for fully voiced stops, where voicing starts before the stop is released,as found in most non-English languages).Lan-guages classify their stops largely based on Voice Onset Time,and it’s an excellent,more gradient empirical measure of the"voiced/voiceless"phonological distinction.
Measuring Voice Onset Time(VOT)is very easy to do in Praat,as it’s just a duration measure-ment between two set points,the release of the stop and the start of voicing.
1.Find the stop release
13
2.Find the start of voicing
3.Select the span between these two points
4.Read the duration of the selection(in seconds)from the duration bar along the bottom
of the Editor window OR
5.Editor→Query→Get selection length and read your selection in the info window
6.If the start of voicing came before the stop release,the VOT is negative.Otherwise,the
VOT is positive.
In general,voiced sounds(in languages other than English)will have a VOT which is negative, voiceless unaspirated sounds will have a VOT which is around0,and aspirated sounds will have a positive VOT.
6.4Examining and measuring F0/Pitch
F0and Pitch can be measured in a number of ways in Praat,more and less reliably.
6.4.1Measuring F0from a single cycle
The surest way to get an accurate F0for a single cycle is to open the?le in the Editor window, then:
1.Zoom in to the point where you can see individual cycles in the sound?le
2.Select one complete cycle,as accurately as possible,thus,giving Praat the period in
seconds(t)
3.Praat will calculate the frequency of the sound in Hertz in the top bar,giving it in the
format(___/s).Use the zoom sel button to zoom in if you can’t see the frequency readout.2
6.4.2Viewing Pitch via a narrowband spectrogram
The most reliable way of getting a sense of the pitch through the course of the word in Praat is by examining a narrowband spectrogram with a reduced visible range(0-400Hz for speech). This can be done by editing the spectrogram settings as described in Section6.2.The contours of the harmonics will accurately represent the pitch contours of the voice during the word,and 2This is going to be an accurate number(so long as you gave an accurate period),but you’re welcome to
,where t is the period in seconds
calculate it yourself to make sure.f=1
t
14
doing this will give you a sense of the contour before using the Praat pitch tracker for more precise measurement.
6.4.3Using Praat’s Pitch Tracking
Praat does have the ability to provide a pitch track in the editor window.To enable the pitch track in the Editor window:
Editor→Pitch→Show Pitch
At this point,a blue line will be placed on top of the spectrogram representing the pitch,where Praat can?nd it.Once the pitch track is placed,you can use the cursor to check the pitch at any given point in the word.Just place the cursor and look for the middle blue number on the right side of the window.You can also place your cursor at a given point in the?le and Editor→Pitch→Get Pitch.Running Editor→Pitch→Get Pitch when a chunk of the sound is collected will return the average pitch during that selection.
6.4.4Improving Pitch tracking by changing the Pitch Settings
It’s worth noting,though,that Praat’s pitch tracking can be quite?nicky.You will often see it jump up and down,doubling and halving the actual F0,and in many cases,especially where the speaker is at all creaky,the pitch track will drop out altogether.This does not represent any speci?c failing of the software,but instead,comes from the variability and noise inherent in actual phonetic data.Part of the strength of Praat’s approach is that you as a user can help Praat improve its pitch tracking for a given?le or speaker by changing some of the Pitch settings.
So,in order to do any serious research using the pitch track,and to avoid some of the problems discussed above,you may need to adjust some of the pitch settings,to help Praat’s pitch tracker better re?ect the speaker’s voice.To do so:
Editor→Pitch→Pitch Settings...
Then adjust the settings as follows:
?Pitch Range(Hz)
–Set the pitch range to a reasonable range(50-400for general usage,going much
higher for song or children’s voices).If you have a good idea what the speaker’s
actual range is(taken from a narrowband spectrogram,for instance),set the min-
imum to just under the speaker’s lowest F0and the maximum to just over their
highest pitch excursion.Changing the minimum pitch will also change the smooth-
ing of the intensity line,when displayed.
15
?Unit
–Here you can choose your unit of choice for the display of the speaker’s F0,ranging
from Hertz to Semitones to auditory-scaled measures like logHz or mel ?Method
–You’ll usually keep this parameter set to"autocorrelation",but you can switch to
"cross-correlation"to see if it improves your pitch track.
You may want to tweak the advanced settings as well.To do so:
Editor→Pitch→Advanced Pitch Settings...
Then adjust the settings as follows:
?Silence threshold
–This is the amplitude threshold for what Praat considers to be speech,relative to
the peak amplitude in a?le.If Praat can’t?nd ANY pitch in a quiet?le,adjust this
setting.
?Voicing threshold
–Praat uses this value in its algorithm to help it decide whether voicing is present.If
Praat is?nding voiceless portions of the word as voiced,raise this number.If Praat
isn’t detecting voicing that’s there in the signal,lower it.This is mostly important
when working with data with either very large or very small amounts of background
noise,and can often be left alone.
?Octave Jump cost
–Changing this value affects the algorithm’s decision about whether a jump in F0is
https://www.sodocs.net/doc/e517798544.html,rger values disfavor abrupt changes in F0.Increase this number if
you are getting pitch-doubling,decrease it if you are failing to track actual rapid
changes in F0.
?Voiced/unvoiced cost
–Increasing this number will make Praat more reluctant to claim a transition be-
tween voicing and voicelessness.Turn this up if your pitch track is cutting in and
out more than is reasonable.
16
Danger!Praat’s pitch tracking is a good way to get a rough idea of what’s going on with the speaker’s F0,but relying on it to give you sane measures is not wise,especially in scripts.Make sure you sanity-check any measures which seem unreasonable against single-cycle F0measurements or against harmonic frequencies,and that you throw out anything completely ridiculous.
6.4.5Scripting only:Creating a Pitch Object
If you’re not scripting,disregard this section.Working from a pitch object will win you nothing in a human-centric work?ow.
When scripting,you may want to create a Pitch object(select the sound,then Objects→Periodicity→To Pitch,specifying the proper range)so that you don’t need to open the editor to measure pitch.Once a pitch object is created,you can instead select the Pitch object and run Objects→Query→Get value at time...to?nd the pitch at whatever time you’d like.Pitch in a pitch object is calculated in the same way as in the Editor window,so the same disclaimers apply.
6.4.6Getting Maximum,Minimum,and Average pitch for a section of speech
This is easy.
1.Select the portion of the sound for which you’d like the Maximum,Minimum or Average
Pitch
2.Select the proper command for your task from the Editor→Pitch menu.
Note that Editor→Pulses→Voice Report will give this information as well.
6.5Measuring Pulses,Jitter,Shimmer,and Harmonics-to-noise ratio
As a part of its pitch-handling system,Praat includes the ability to?nd individual glottal pulses in a signal and to analyze the pulses as part of more complex analyses.To view these pulses, Editor→Pulses→Show Pulses,and they’ll then display on top of the waveform in your?le.Al-though the pulses themselves are mostly only useful in scripting,the Editor→Pulses menu con-tains one of the more useful commands in the program:Editor→Pulses→Voice Report
To use this command,simply select a voiced section of the sound,then Editor→Pulses→Voice Report.An information window will then pop up providing you with a variety of useful mea-sures.In addition to maximum and minimum pitch(with additional statistics),you will also
17
be given the jitter,shimmer,harmonics-to-noise ratio(HNR),and the noise-to-harmonics ratios for the selected portion of the sound.
Jitter is a measure of the periodic deviation in the voice signal,or the pitch perturbation of the signal.Put differently,each cycle of speech with a given F0should,in a perfect world,have the same period.The jitter in a person’s voice is how much one period differs from the next in the speech signal.This is a useful measure in speech pathology,as pathological voices will often have a higher jitter than healthy voices.
Shimmer(amplitude perturbation)is similar to jitter,but instead of looking at periodicity,it measures the difference in amplitude from cycle to cycle.Once again,this is a useful measure in speech pathology,as pathological voices will often have a higher shimmer than healthy voices.
Harmonics-to-noise ratio(HNR)and Noise-to-harmonics ratio are both measures of the amount of periodic noise compared to the amount of irregular,aperiodic noise in the voicing signal. Because the aperiodic noise often represents frication in the vocal tract,the HNR will go down signi?cantly with hoarse or breathy speech,and other laryngeal pathologies will lower the HNR further still.
6.6Measuring Formants
Praat has several methods of built in formant measurement.Of course,the easiest way to examine formant heights is by simply looking at a broad-band spectrogram and using the cursor to?nd,roughly their frequencies.However,"eyeballing it"won’t pass scienti?c muster, and is more time consuming than using Praat’s built-in LPC algorithms as a tool to help to?nd them.
6.6.1Using the Formant tools in the Editor window
When you open a sound?le in the Editor window,you can choose to have Praat calculate and display where it thinks that the vowel formants are(Editor→Formants→Show Formants). This will overlay a series of red dots onto the image which represent peaks in the series of LPCs which Praat has run.
This formant track can be queried at any time in a variety of ways,all accessed through the Editor→Formants menu.If you’re interested in a single formant’s height,you can place the cursor where you want a measurement and choose Editor→Formants→Get formant...,but it’s often more ef?cient to use Editor→Formants→Formant Listing,which will give you heights for F1,F2,F3and F4,along with the timepoint at which the measures were taken.
18
Figure4:Praat’s Formant Settings Window
In addition,if you’re interested in formant bandwidth,bandwidth for the?rst four formants can be taken using the Editor→Formants→Get___Bandwidth commands.
For hand measurement,using Editor→Formants→Formant Listing and sanity-checking by visually inspecting the formants on the Spectrogram will usually produce reasonable results, but there are ways to improve Praat’s formant-picking performance for a given speaker.
6.6.2Improving Formant Finding results
For most speakers,the default settings will suf?ce,but if you?nd Praat to be struggling with "missing"or the addition of extra formants,you’ll likely?nd that that particular speaker’s formants are more effectively measured if you make some tweaks to Praat’s Formant Settings, helping the computer with its task.
To apply any of these changes,you’ll want to open the Formant Settings window(see Fig-ure4):
Editor→Formants→Formant Settings....
Praat has to make a series of guesses about how many formants it will?nd,and how spread out those formants will be.We typically will assume that speakers will have one formant per 1000Hz,and thus,that there will be5formants in the5000Hz we usually worry about for speech research.Thus,"5"is the default setting for Number of Formants,and the highest we’ll look for formants(the Maximum Formant)is5000Hz by default.
Usually,you’ll only need to adjust the Number of Formants.Although5formants is a good
19
baseline,if Praat is ?nding formants where there are none (between two actual formants,usually),you should lower this value down to 4or 3.If Praat is ?nding too few formants (missing F 2and labeling F 3as F 2,for instance),you’ll want to raise this number up to 6.
If you’re working with a child,a person of unusually small stature,or somebody with an otherwise tiny vocal tract,you may ?nd that the Praat is ?nding non-existent formants between the speaker’s F 1and F 2,and missing the speaker’s higher formants (F 3and F 4)altogether.In this situation,you’d want to increase the Maximum Formant (Hz)value to tell Praat to search a bit higher up in the spectrum for formants,and perhaps lower the number of formants it’s searching for.
Realize,though,that Praat can always ?nd more peaks,and there are often small peaks not perceptible to humans which may still have an acoustical relevance.When there are "too many formants",Praat is not necessarily ?nding formants which "aren’t there",but is ?nding additional peaks which,although present,aren’t the F 1,F 2and F 3peaks which we as linguists are chie?y interested in.When there are "too few",Praat is just giving you only the most prominent peaks that you’ve asked for.The results of Praat’s formant tracker,in reality ,are largely determined by what you’re asking it to ?nd,and this settings adjustment be done with a mind to what you’re actually interested in.
Dot size (mm)simply controls how large the red dots in the formant display are.Although changing this can be useful if the track obscures the spectrogram,this will have no effect on your measurements.
Just remember,these settings persist even once you’ve closed Praat,so if you make adjustments here,you’ll want to return these settings to the defaults when you’ve ?nished with your odd
speaker.
Danger!
No matter your settings,Praat will happily ?nd you formants even in fricative noise
or silence,and because it doesn’t know how many formants it should be searching
for where,it commonly merges F 1and F 2for high back vowels (where they’re close
together).In addition,Praat will often have issues ?nding a single timepoint,so if
you’re getting an unusual measurement,a timepoint shortly before or after may be
more accurate.Always sanity-check your measurements,make sure you know what
you’re measuring,and during automated measurement,always run the results by a
trained human ?rst!20
视频分析系统建设方案 1.1.1系统概述 本次视频分析系统共涉及摄像机11个、人流量监控点5个。为保证系统一致性确保视频分析效果、系统兼容性开发量等多方面因素，考虑采用原有品牌设备。 1.1.2系统架构 1.1.3系统功能 1.1.3.1经典设计，奠定智能分析基础 感知型人员卡口摄像机，机身采用铝质金属外壳，边侧设计有散热风扇，外观为黑白烤漆配色，在适配的镜头组合下，样式也很可人美观。其采用200 万1/1.9" 大靶面逐行扫描图像传感器，成像清晰，中心清晰度达到1100TVL;同时在良好的算法功能支持下，能实现彩色0.002Lux(@F1.2，AGC ON)、黑白0.0002 Lux(@F1.2，AGC ON)的星光级超低照度效果;另具有宽动态等功能，满足各类应用场景的使用需求。 1.1.3.2平台化运作，深挖智能深度
该摄像机的前端智能化、数据结构化设计，无需工控机二次处理，语义信息直接汇聚到后端管理平台数据库中，同时可以采用云架构组网，实现语义信息、图片信息、视频信息的云存储、云调用。进入公安实战平台的人员卡口系统可看到，其平台能提供如下功能： 1.1.3. 2.1?人员实时监控： 与摄像机B/S客户端的功能基本一致，不过可以同时监控3支设备的抓拍情况并任意调阅网内的所有人员卡口摄像机; 1.1.3. 2.2?人员卡口布控： 在布控时间段内对布控点位出现的所有人员与数据库中的嫌疑人进行实时比对，同时可以对比对的相似度进行设置，以提高布控精度。现场模拟测试中，当卡口布控场所出现与数据库中“嫌疑人”相似的人员出现时，系统即报警，测试准确率也为100%(相似度设置为60%); 1.1.3. 2.3?人脸库以图搜图功能： 除了卡口布控，也可通过手动输入人员图片，对数据库中与之相似的卡口抓拍人物头像进行1:N比对，测试响应速度<1s，检索所得图片以相似度由高到低排列，测试中相似度靠前的几张图片即为被检索人，检索快而准; 1.1.3. 2.4?综合语义分析： 这也是该公安实战平台充分利用感知摄像机语义信息进行大数据分析和应用的体现，可以通过布控点位、类型、摄像机提供的语义信息(通过时间、行走方向等)以及后台二次分析所得的民族、年龄、性别等语义选项执行检索;同时支持中英文模糊检索，实测也显现出检索反应快速。 1.1.3.3前端数据结构化，专一感知世界 以“重新定义”作为感知型摄像机的标签，当时很多人都不理解为何要用“重新定义”一词，其实这正展示了对智能分析的重新理解。 1.1.3.3.1从数据源头识别 该摄像机从非压缩数字信号源中提取原滋原味的特征信息(提取后才压缩传输视频流)，保证了分析及识别结果的准确性。 1.1.3.3.2形成语义信息
用Praat标注操作步骤 一、打开praat软件，只保留praat objects窗口 二、点击read→点击read from file→选择录音文件→点击右手边的Annotate选项→点击To TextGrid选项→在弹出的Sound：To TextGrid界面第一个对话框输入1，第二个清除干净不用管→点击OK→按住Ctrl键，同时选中wav文件和textgrid文件，点击右侧的Edit，出现标注页面，开始进行标注 三、标注常用操作指令 1.播放/暂停：Tab键 2.放大/缩小：界面左下角：all全屏显示；in逐步放大；out逐 步缩小；sel选中部分全屏显示 3.选中音频：在语音波形上拖动鼠标 4.拖动音频：拖动标注界面最下的滑动条 5.生成切割线：在语音波形上用鼠标点击需切割处，即出现一 条红色虚线，同时该红色虚线与每个标注层的相交处有一个空心圆圈，点击空心圆圈，即可生成切割线（快捷键：Enter） 6.移动切割线：鼠标点住要移动的切割线，左右拖动 7.删除切割线：点击界面左上顶部Boundary选项→选择 Remove，即可删除（快捷键Alt+Backspace） 8.保存：Ctrl+S 9.查看秒数：在标注层下面，滑动条上面，有三个显示依次为： 每个切割片切割秒数、屏显秒数、整条音频秒数
10.做完之后删除记录：选中文件之后，点击左下角的Remove 四、对一个标注过的录音文件（可能没标完、可能已标完）进行修改或查看 1.双击标注软件praat，弹出三个界面，只保留praat objects即可，其他关闭 2.从praat objects菜单中点open→read from file，分别载入录音wav文件和TextGrid标注文件，按住Ctrl键同时选中这两个文件，点击右侧的Edit即可在弹出的标注界面里修改
视频质量分析软件简介1. 视频图像类别
2. 参数设置 2.1阈值及检测项设置 对正常显示，但画质欠佳的各检查项设置阈值。 阈值归一化为[0 1]区间，其中0表示敏感度最低,也即只选出相似度较高的干扰项，1表示敏感度最高，即把可能存在的干扰项全部选出。 检测项设置用于强制检测非常规的干扰，包括字迹和横向移位两项。当用户需要或特定应用场景下会出现该类故障时，可选中这两项。 2.2位置参数设置 不同的监控系统中，往往在图像上方或下方添加时间、位置、状态等文字属性信息，这类信息可能会对图像质量分析产生干扰。位置参数设置用于设置这类文字属性信息的位置，以相对于原始图像百分比的形式给出，使识别结果更稳定。
3. 库函数接口 3.1 接口简介 定义类CDigitVerify及接口函数void CDigitVerify::ProcessMain(int* anResult,unsignedchar* abyImageColor, int nHeight,int nWidth, float* afParas, float* afBounds)，其中a nResult为输出结果,abyImageColor为输入RGB图像数据, nHeight为输入图像高度, nWidth为输入图像长度, afParas 为输入阈值, a fBounds为输入上下边界。 3.2 例程 //读取图象 m_Image.Destroy(); HRESULT hResult = m_Image.Load(m_StrFileName); if (hResult!=S_OK){ // 未成功读取 nResult[0]=0; m_ListView.SetItemText(nItem, 3, "无法读取"); continue; } int nWidth= m_Image.GetWidth(); /// 长度 int nHeight= m_Image.GetHeight(); /// 高度 int nBitCount=m_Image.GetBPP(); /// 获取每象素的数据位数 int nRowByte =m_Image.GetPitch(); /// 相邻两行象素首地址间的间隔,可正可负
智能视频分析系统
目录 一、项目背景及建设目标 (3) 1.1 项目背景 (3) 1.2 技术优势 (4) 二、厂区智能视频分析整体设计方案 (5) 2.1传统对射系统与智能视频分析系统比较 (5) 2.2厂房周界入侵报警系统 (6) 2.2.1 周界入侵检测 (7) 2.2.2 周界警戒线警戒区预警 (8) 2.3厂房仓库物资看护 (8) 2.3.1 可疑人员接近仓库提醒 (8) 2.3.2 仓库物品看护 (9) 2.3.3 夜间停车场、厂区内部、附近可疑逗留检测 (9) 2.4夜间厂区办公楼内可疑人员检测 (10) 2.5生产车间危险区域或者夜间下班后人员检测 (10) 2.6系统拓扑结构 (11)
一、项目背景及建设目标 1.1 项目背景 慧视科技智能视频分析系统是以软件的形式实现智能视频分析功能，拥有自主的软件知识产权，可满足各行业的需要，也满足各厂家设备的接入，同时可以与各种监控平台进行二次对接。传统报警设备的误报多漏报多操作复杂不直观已经成为行业共识，且传统的视频监控系统数量庞大画面单一，工作人员很难从视频中发现问题，往往更多用于事后取证，智能图像分析通过图像中目标的识别和规则运用来进行预警，报警速度快且精确度高，可辅助工作人员从繁琐重复的工作中解放出来，真正体现科技为人服务的理念。 国内现有厂房的视频监控系统主要由摄像机、光缆、矩阵、硬盘录像机和电视墙等组成。由于视频监控图像数量大，内容枯燥，现有系统即使配备值班人员，在大多数情况下仍处于无人观看的状态下。当犯罪事件发生时，从硬盘录像机中调取录像回放、取证变成系统主要的价值之一。即使值班人员在岗，由于人的生理特点，不可能长时间有效观察多路图像，很可能造成遗漏可疑事件，对安全形式产生错误判断。 智能视频监控技术可以理解为用计算机来帮助值班人员"看"监控录像。现代计算机的高可靠性可以提供24小时不间断地保护。从根本上杜绝由于人员疲劳造成的遗漏问题。同时也可以防止出现监控人员内外勾结的可能性。
第1部分:语音的基础知识 1.1语音的基础知识,以及如何使用Praat语音学 在本教程中,您将得知声学的基本元素和语音研究。它被分为五个模块:基本的声学1和2,声譜图、LPC和基音同步叠加,反过来，他们也将被分成不同的段落，详见下列索引。 每个模块是由理论部分和的实践部分组成(答案在单独的文件)。通过做习题来检查您理解,因为答案有时也是不准确的，答案有事也是解释的一种互补。每个模块也由实践部分组成,您将学习如何使用Praat。Praat是一个对说话进行分析的开放性资源项目。这个程序是由保罗~Boersma和大卫Weenink开发出来的(阿姆斯特丹大学)。 为了在你的电脑上安装praat软件，进这个网站< http://www.fon.hum.uva.nl/praat > >下载Praat(选择版本用于您的操作系统)。这可能需要花一些时间来完成,要么通过下载自解压可执行文件,双击或者下载ZIP文件，双击打开它。用这种方式, Praat或Praat.exe文件会出现,这是Praat程序本身。它可以非常方便地在您的电脑上生成一个praat软件的快捷方式,例如在桌面(在Windows中,点击Praat。exe与鼠标右键> >按钮发送到桌面,或者点击右键单击桌面快捷方式> > >新浏览找到Praat文件)。还有一个Praat用户列表(< https://www.sodocs.net/doc/e517798544.html,/group/praat-users/ >),那里的人们分享他们的知识张贴问题和答案。 音频文件,我们从是下载通过 http://www.let.uu.nl/ ~ Hugo.Quene /个人/ onderwijs / sprekenenverstaan / practicum20070725.zip < http://www.let.uu.nl/%7EHugo.Quene/personal/onderwijs/spr ekenenverstaan/practicum20070725.zip > 或从 http://www.let.uu.nl/ ~ audiufon /数据/ pracfon1.html < http://www.let.uu.nl/%7Eaudiufon/data/pracfon1.html >。1.2声波
如何使用Praat １.下載練習檔 在你的電腦裡新增一個資料夾，給這個資料夾取個名字，比如說「語概」。在教學平台「課程文件」可以看到兩個檔案，一個叫「阿扁錄音.wav」的音檔、一個叫「阿扁錄音.txt」的標記檔。請將游標分別移到這兩個超連結，按滑鼠右鍵，然後選「另存目標」，存到剛才建立的檔案夾裡。 ２.下載及安裝 如果你還沒安裝過Praat，可以點選底下名叫「Praat: doing phonetics by computer」的超連結，連結到該網站，在「Download Praat:」項目下選擇你所用的版本，如果你選的是「Windows」，那麼進入下一頁之後，只要點選 「praat5342_win32.zip」或「praat5342_win64.zip」就可以下載到你的電來按裝。 ３.啟動及設定。 安裝完成後，會有一個的圖示，點兩下就可以啟動。你的電腦上會出現兩個大框，將游標移到那個叫「Praat objects」的方框上，點「Praat」選項，會出現一個清單，再將將游標移到「Preferences」，它的右側又會出現一個清單 ，點選「Text writing preference」，會出現另一個框
再點「UTF-8」，然後按「OK」。 ４.用Praat開啟檔案 在「Praat objects」裡點「Open」，會出現一個選單， 再點「Open long sound file」，開啟下載的「阿扁錄音.wav」檔。再點「Open」，
再點「Read from file」，開啟下載的「阿扁錄音.txt」檔。這時「Praat objects」裡會有兩個項目，按住鍵盤的「Ctrl」，用滑鼠分別點，可以同時選取。當兩個項目都變成藍底時，按旁邊的「View & Edit」鍵，會出現一個名叫「阿扁錄音」的框，這是聽寫工作主要的工作區。你現在可以點這些框邊的叉叉，試著把它們關掉、開啟幾次。 ５.用Editor聽音 在「Praat objects」裡選取檔案、按下「View & Edit」鍵之後，出現的那個框叫作「Editor」。如果你已經開啟了練習用的兩個檔，那麼「Editor」左上角就會顯示你所編輯的檔案叫作「TextGrid阿扁錄音」。以下的說明，你最好搭配著「Editor」，一面看一面作。 「Editor」的中心分成三個橫欄，上兩欄是聲音圖，第三欄是文字欄。試試看在聲音圖的任何地方用滑鼠點點看，有沒有發現滑鼠點到的地方就會跑出一條直立的虛線，而第三橫欄的某一格就會變黃？這些寬窄不一的格子很重要，因為聽寫的時候要用到它。
五家主流视频监控平台介绍与对比分析 目录 一、视频监控软件平台发展与趋势 (3) 1、第一代：孤岛式平台 (3) 2、第二代：联网型平台 (4) 3、第三代：运营级平台 (6) 4、第四代：面向智慧交通的云平台 (6) 二、"联网型"视频监控软件平台比对 (7) 1、视频监控软件平台厂家概况 (7) 2、主流厂商介绍 (8) 2.1高新兴科技集团股份有限公司 (8) 2.2浙江大华技术股份有限公司 (9) 2.3海康威视数字技术股份有限公司 (9) 2.4东方网力科技股份有限公司 (10) 2.5华平信息技术股份有限公司 (11) 3、比对分析 (11) 3.1专注领域 (11) 3.2地域倾向 (13) 3.3发展分化 (14) 4、平台软件介绍 (14) 4.1高新兴：C3M-Video2.0 视频综合管理平台 (14) 4.2大华：DSS-T综合监控管理平台-智能交通软件平台 (19) 4.3海康：iVMS-8600智能交通综合管控平台 (22) 4.4东方网力：PVG网络视频管理平台 (28) 4.5华平：AVCON监控平台软件 (30) 三、主流平台对交通业务的支持 (36) 四、平台软件比对分析 (40) 1、系统架构 (40) 2、可靠性 (41) 3、兼容性 (42) 4、扩展性 (43) 5、可管理性 (44) 五、对比总结以及选型建议 (45) 1、对比总结 (45) 2、选型建议 (46)
如何选择合适的交通视频监控平台 --五家主流视频监控平台介绍与对比分析 视频监控一直都是交通管理部门的重要业务组成部分，通过安装在主要道路、公交站场的摄像机，管理部门可以实时监控到现场的情况，对历史录像下载回放以还原事件的真实经过，对突发事件进行有效的应急指挥调度。因此，视频监控对保障交通通行、维护正常的交通秩序起到重要作用。 由于视频监控的实时性和可溯性难以替代，在智能交通的规划和建设中，视频监控是高速公路信息化的重要组成部分。但是，在实际的运营中，传统的视频监控软件平台已难以跟上智能交通快速的发展要求，主要体现在几个方面： 1、由于各种原因，存在多个视频监控平台，平台与平台之间缺乏信息共享，数据孤立，难以充分发挥设备和系统的最大效用。 2、平台缺乏伸缩性，无法应对前端监控点的快速扩容。 3、平台缺乏开放性，平台之间无法级联，无法兼容不同品牌的前端设备。对一些新接入的设备（如车载视频）需重新开发。 4、平台多为通用版的产品，没有在智能交通的大框架下规划和设计，无法成为新建的智能交通系统中的一员。 由于先天的局限性，原有的视频监控平台只能由原厂家修修补补，勉强应对新业务的需求。这种"头痛医头脚痛医脚"的方式严重影响了系统的稳定性和可靠性，为了维系系统的稳定性，开发了一些所谓的性能检测系统等，这些系统又额外地增加了原系统的负荷，造成了新的不稳定因素。周而复始，恶性循坏，最终导致整个系统轰然倒塌，业务停止，混乱不堪。 解决这些问题的根本方法就是用开放性、前瞻性的设计理念指导平台的规划，用高伸缩
基于智能视频分析的监控平台建设方案 随着国家经济的提高，城市和城市化进程在不断的发展，各种社会矛盾和暴力事件逐渐增多，政府和相关部分对加强城市各地联网型监控系统越来越重视，当前城市和小区监控系统建设使用监控录像存储，事件发生后调取查阅的方式，这种方式在一定程度上满足了社会的需求，但是无法避免事态趋于恶化，在此背景下，具有智能视频行为分析的监控平台建设就显得尤为重要。 智能视频技术让安全警卫部门能通过摄像机实时自动“发现警情”并主动“分析”视野中的监视目标，同时判断出这些被监视目标的行为是否存在安全威胁，对已经出现或将要出现的安全威胁，及时向安全防卫人员通过文字信息、声音、快照等发出警报，极大地避免工作人员因倦怠、脱岗等因素造成情况误报和不报，切实提高监控区域的安全防范能力。 现有各大监控系统厂商和信息化科技公司都研发出大量的智能视频分析软件，可以分为两大类，基于嵌入式DSP 智能分析系统和基于计算机末端处理的智能分析系统。 一．基于嵌入式DSP的处理优点
1、DSP方式可以使得视觉分析技术采用分布式的架构方式。在此方式下，视觉分析单元一般位于视觉采集设备附近（摄像机或编码器），这样，可以有选择的设置系统，让系统只有当报警发生的时候才传输视觉到控制中心或存储中心，相对于计算机末端处理方式，大大节省的网络负担及存储空间。 2、DSP方式下视觉分析单元一般位于视觉采集设备附近（摄像机或编码器），此方式可以使得视觉分析单元直接对原始或最接近原始的图象进行分析，而后端计算机方式，计算机器得到的图象经过网络编码传输后已经丢失了部分信息，因此精确度难免下降。 3、视觉分析是复杂的过程，需要占用大量的系统计算资源，因此计算机方式可以同时进行分析的视觉路数非常有限，而DSP方式没有此限制。 二．在对比上述两种处理模式的优缺点基础上，提出基于DSP嵌入式处理和末端计算机处理两种系统结构.
Using Praat for Linguistic Research Will Styler University of Colorado at Boulder Phonetics Lab Document Version:1.3.6 Last Update:October2,2013 Important! This document will be continually updated and improved.Download the latest version at: https://www.sodocs.net/doc/e517798544.html,/praat Using Praat for Linguistic Research by Will Styler is licensed under a Creative Commons Attribution-ShareAlike3.0Unported License.For more information on the speci?cs of this license,go to https://www.sodocs.net/doc/e517798544.html,/licenses/by-sa/3.0/. 1
Contents 1Version History4 2Introduction5 2.1Versions (5) 2.2Other Resources (5) 2.3Getting and Installing Praat (6) 3About Praat6 3.1Praat Windows (6) 4Recording Sounds8 5Opening and Saving Files8 5.1Opening Files (8) 5.1.1Working with longer sound?les (9) 5.2Saving Files (9) 6Phonetic Measurement and Analysis in Praat9 6.1Working with Praat Waveforms and Spectrograms (9) 6.1.1Pulling out a smaller section of the?le for analysis (11) 6.2Adjusting the Spectrogram settings (11) 6.2.1Narrowband vs.Broadband Spectrograms (12) 6.3Measuring Duration (13) 6.3.1Measuring Voice Onset Time(VOT) (13) 6.4Examining and measuring F0/Pitch (14) 6.4.1Measuring F0from a single cycle (14) 6.4.2Viewing Pitch via a narrowband spectrogram (14) 6.4.3Using Praat’s Pitch Tracking (15) 6.4.4Improving Pitch tracking by changing the Pitch Settings (15) 6.4.5Scripting only:Creating a Pitch Object (17) 6.4.6Getting Maximum,Minimum,and Average pitch for a section of speech.17 6.5Measuring Pulses,Jitter,Shimmer,and Harmonics-to-noise ratio (17) 6.6Measuring Formants (18) 6.6.1Using the Formant tools in the Editor window (18) 6.6.2Improving Formant Finding results (19) 6.6.3Scripting Only:Formant Objects (21) 6.7Measuring Intensity/Amplitude (21) 6.8Working with Spectra (22) 6.9Taking a spectral slice (22) 6.10Measuring Harmonic Amplitude,Frequency (23) 2
智能视频分析系统解决方案 1.1 系统概述 智能视频（Intelligent Video）技术源自计算机视觉（Computer Vision）与人工智能（Artificial Intelligent）的研究，其发展目标是在图像与事件描述之间建立一种映射关系，使计算机从纷繁的视频图像中分辩、识别出关键目标物体。这一研究应用于安防视频监控系统，将能借助计算机强大的数据处理能力过滤掉图像中无用的或干扰信息，自动分析、抽取视频源中的关键有用信息，从而使传统监控系统中的摄像机成为人的眼睛，使“智能视频分析”计算机成为人的大脑，并具有更为“聪明”的学习思考方式。这一根本性的改变，可极大地发挥与拓展视频监控系统的作用与能力，使监控系统具有更高的智能化，大幅度节省资源与人员配置，同时必将全面提升安全防范工作的效率。因此，智能视频监控不仅仅是一种图像数字化监控分析技术，而是代表着一种更为高端的数字视频网络监控应用。 智能视频分析包含视频诊断、视频分析和视频增强等，它们各自又包含了大量的功能算法，比如清晰度检测、视频干扰检测、亮度色度检测、PTZ（云台）控制功能检测，以及视频丢失、镜头遮挡、镜头喷涂、非正常抖动等检测都属于视频诊断内容，而视频分析算法则包含区域入侵、绊线检测、遗留遗失检测、方向检测、人群计数、徘徊检测、流量统计、区域稠密度统计、人脸识别、车牌识别、烟火烟雾检测、自动 PTZ 跟踪等功能，视频图像增强则包括稳像、去雾、去噪、全景拼接等算法。由此组合衍生出的算法种类又有很多，应用方式也千变万化，所以智能视频分析的应用范围很广。 在以往的视频监控系统中，操作人员盯着屏幕电视墙超过 10 分钟后将漏掉90％的视频信息，而使视频监控工作失去意义。随着社会发展，视频监控被越来越广泛地应用到各行各业中，摄像机数量越来越庞大，这给传统的视频监控带来严峻的挑战。针对行业发展推出智能视频分析系统，主要解决以下问题：一个是将安防操作人员从繁杂而枯燥的“盯屏幕”任务解脱出来，由机器来完成分析识别工作；另外一个是为在海量的视频数据中快速搜索到想要找的的图象。 1.2 系统组成 智能视频分析系统以数字化、网络化视频监控为基础，用户可以设置某些特定的规则，系统识别不同的物体，同时识别目标行为是否符合这些规则，一旦发现监控画面中的异常情况，系统能够以最快和最佳的方式发出警报并提供有用信息，从而能够更加有效的协助安全人员处理危机，最大限度的降低误报和漏报现象。智能视频分析是在传统的监控系统中，加入智能视频技术，在整个系统中，系统分布图如下：系统分布图 智能视频分析系统是由模块化的设计组合而成，整个系统主要由视频源、视频智能处理引擎、管理终端组成。 视频源的主要功能是通过前端摄像机对关键区域实时视频进行采集，通过网络、光纤传回监控中心。 智能视频处理引擎是一台可以进行智能视频分析的服务器，它能够按照设备内预设的规则对视频源采集到的实时视频进行实时的智能分析，当收集到的视频中满足设备内预存的规则时，引擎向外发出告警。
智能视频分析系统简介 2015年8月10日 1
目录 引言： (3) 一、系统概述 (3) 二、技术原理 (4) 2.1视频分析方法 (4) 2.2视频分析过程 (4) 三、系统架构 (5) 3.1逻辑结构 (5) 3.2物理架构 (5) 四、系统优势 (6) 4.1算法先进、稳定、抗干扰能力强 (6) 4.2灵活制定监控策略 (7) 4.37×24全天候可靠监控 (7) 五、应用案例 (7) 5.1云南农信社智能视频分析系统项目 ·································错误！未定义书签。 2
引言： 智能视频分析系统是一种涉及图像处理、模式识别、人工智能等多个领域的智能视频分析产品。它能够对视频区域内出现的警戒区警戒线 闯入、物品遗留或丢失、逆行、人群密度异常等异常情况进行分析，及 时发出告警信息。 一、系统概述 随着计算机和网络技术的发展，视频监控系统逐渐由模拟时代过渡到数字时代：包括视频监控信号的数字化采集、压缩、传输、存储等都 实现了数字，但是智能化程度还很低。近几年，随着计算机性能的不断 提高，利用计算视觉技术对监控系统中的视频信息进行分析，从而使视 频监控系统向智能化方向发展成为主流趋势。 智能视频分析系统能够对视频区域内出现的运动目标自动识别出目标类型并跟踪，对目标进行标记并画出目标运动轨迹，能够同时监测同 一场景里多个目标，可以根据防范目标的特点进行灵活设置；它能够适 应不同的环境变化，包括光照、四季、昼夜、晴雨等，并能够很好地抗 摄像头抖动。其改变了以往视频“被动”监控的状态，不仅仅局限于提 供视频画面，而且能主动对视频信息进行智能分析，识别和区分物体， 可自定义事件类型，一旦发现异常情况或者突发事件能及时的发出警报，其在安防领域的应用必然有助于克服了人力疲惫的局限性，从而更加有 效地协助安全人员处理突发事件。 3
视频直播系统解决方案 视频直播简介 视频直播，准确来讲是音视频直播，即将音视频信号压缩成数字信号，并通过IP网络进行传输的一种流媒体应用。视频直播和视频点播的区别在于，视频直播讲求信息的实时性广播，视频点播强调信息的娱乐性和个性化；视频直播和视频会议的区别在于，视频直播讲求的是信息以点对面的形式实时传播，视频会议突出的是几点之间的交流与协作。 视频直播应用前景 视频直播在不同的行业的应用前景非常明确、宽泛，教育行业的多媒体教学、远程教育、校园电视台、考场监控等，军队行业的远程军事教育、军事演习的网络直播等，医疗行业的临床教学、专家会诊、手术直播等，政府、企业的会议活动、内部培训、产品展示、在线招聘、视频监控等，还有在线路演、拍卖、竞标等等。典型的行业应用有： · 基于IP网络的远程教学、培训 · 集团式商业机构的远程巡查、监控 · 信息服务的网站的会员制视频直播服务 · 政府企业搭建自己的网络视讯平台，进行会议和其他活动的直播 世纪葵花视频直播系统软件 世纪葵花音视频直播系统是高质量的Mpeg4音视频直播软件，能够提供基于局域网、城域网、广域网以及卫星网的音视频直播解决方案。 (1)、世纪葵花音视频直播系统介绍 世纪葵花音视频直播系统是北京世纪葵花数字传媒技术××公
司自主研制开发的音视频直播系统，该系统采用分布式的理念，结合世纪葵花一贯的开放式设计原则，单台普通服务器的性能可支持的并发用户数达5000人以上，并支持无极扩展，自动平衡，系统可自动根据用户的情况平衡负载，达到最大限度的用户连接支持。系统支持B/S构架，使用和维护都非常简单，服务器支持Web方式的管理和配置，极大的节约了维护成本。 （2）、音视频直播系统功能特点介绍 · 支持在广域网、城域网和局域网内进行音视频直播，可穿越网关、路由器以及防火墙； · 采用分布式架构，采集、编码、转发、存储和管理灵活配置； · 兼容目前市面上流行的大部分采集设备（支持VFW接口的采集卡，USB采集设备等），支持一机多卡和一卡多路音视 频采集； · 在正常的网络条件下，音视频同步性能很好。音视频同步的误差范围在0.1秒左右，延时可控制在5秒左右或更短 时间。 · 支持服务器集群方式运行以及自动负载均衡，能随着用户数增加而平滑扩容； · 支持B/S、C/S架构，支持远程管理，灵活搭建多种直播的服务模式； · 支持单播、组播、多播等多种分发方式，系统可根据接收用户的不同情况，将各部分灵活搭配，以满足不同传输、 接收方式； · 采用MPEG4标准编解码和RTSP/RTP/RTCP等网络传输协议，支持音、视频码流的无极控制，可根据用户带宽的情况调 整音视频传输，适应局域网、城域网以及广域网等复杂的 网络环境，可同时采集高、中、低三路数据流，同时满足 （56k拨号、ISDN一线通、ADSL宽带等）不同带宽的客户
智能视频分析系统 一、概述 智能视频分析系统是一款基于Windows操作系统的智能视频分析软件。该软件运用计算机视觉技术，通过对连续图像的处理和分析，将运动物体从视频背景中检测、分离出来，再进行筛选、过滤，对运动目标进行分析、跟踪和判断，将符合用户设定规则的事件进行报警和存储，使普通监控系统具备了人工智能的计算机视觉能力。 智能视频分析系统通过对视频里每帧的每一个像素进行自动运算，使监控系统具有自动检测和分析能力，将监控系统从“事后取证”升级到“自动预警-报警”级别。可满足各行业的需要，也满足各厂家设备的接入，同时可以与各种监控平台进行二次对接。 由于连续监视会导致人很容易疲劳，导致整套系统存在一些缺陷。主要表现在：◆人类自身的弱点：有时无法察觉安全威胁，注意能力不可持久。 ◆监控时间：视频监控系统在大部分情况下，视频源的视频画面并没有被安全 人员看到，很可能在这些时间内就有值得注意的异常现象出现。人不可能24小时不间断的盯着监视器，研究表明，人连续监视某个画面22分钟以后，会丢失90%以上的画面信息。 ◆数据分析困难：传统视频监控系统缺乏智能因素，需要24小时完整的录像， 占用大量磁盘存储空间，而且录像数据无法被有效的分类存储，最多只能打上时间标签，数据分析工作变得非常耗时，很难获得全部的相关信息。
◆被动监控：目前的监控系统大部分情况下都仅起到一个“录像”的工作，即 将一段时间内的视频源使用DVR录制下来，异常情况甚至于突发事故已经发生之后，再进行事后的查验，但此时损失和影响已经造成，无法挽回，完全是一种“亡羊补牢”式的“被动监控”。 二、智能分析设计 中恒智能视频分析系统是以软件的形式实现智能视频分析功能，拥有自主的软件知识产权，可满足各行业的需要，也满足各厂家设备的接入，同时可以与各种监控平台进行二次对接。传统报警设备的误报多漏报多操作复杂不直观已经成为行业共识，且传统的视频监控系统数量庞大画面单一，工作人员很难从视频中发现问题，往往更多用于事后取证，智能图像分析通过图像中目标的识别和规则运用来进行预警，报警速度快且精确度高，可辅助工作人员从繁琐重复的工作中解放出来，真正体现科技为人服务的理念。 ?能够看 ?能够想 ?能够说
大规模人群分析预警系统 方 案 书
1、项目背景 在近两年黄金周全国旅游的统计来看，旅游业作为新列入国民战略性经济支柱产业的地位凸显，旅游井喷时代的到来指日可待。随着客流量不断增加，旅游景区和一些旅游城市的管理和服务面临前所未有的挑战：如景区安全、游客管理服务、以及资源保护等，任何一个问题解决不好，都有可能影响一些人群易聚集区域的整体效益和规划目标，并可能出现重大事故。 当前，景区科学管理、旅游城市发展已进入一个新的历史时期，要实现旅游区域新一轮的大发展和新超越，必须依靠科技的手段，着力打造智慧旅游，通过科学管理的建设，实现智慧城市管理，智慧旅游服务、智慧旅游管理、智慧旅游营销的目标，为旅游业的可持续发展奠定扎实基础。 特别在一些人群容易聚集的区域，通过科学的手段能够实时对人群聚集等等异常情况发生预警，并能清晰对当前区域人群做出趋势分析，完成异常事件查询等等为决策者提供决策的帮助。 2．建设目标 项目建设以深度学习高级算法技术为基础，对目标区域进行客流统计，并实时形成数据报表，一方面可对当前人群流量进行统计，在人群高峰期管理方可以此为依据对该区域做出及时部署，避免客流量“超载”，另一方面，客流统计系统应给景区的项目规划和商业活动提供必要的数据支持。
3．大规模人群分析预警系统核心价值大规模人群分析预警系统可以提供监控中每个人群监控点的人群规模数据，将这些数据汇总到数据中心并进行分析汇总，从时间和空间维度对监控区域中人群的分布以图表的形式进行展示。该系统具有以下核心价值： A、大区域大规模人群聚集预警，大规模人群异常状况预警 实时判断人群规模，人群滞留发现，人群异常行为检测，如聚集、逆行、奔跑等。 B、评估营销策略效果，提高销量 通过客流量对比，有效评估所推行的推广活动，对营销和促销回报进行有效评估 通过对历史销量和客流量的对比，可以有效的分析景区服务项目及各项管理策略对流量及销量的影响，进而更好的进行管理决策，提高销量。 通过每天的客流变化规律的统计，可以很好的安排景区工作时间，从而提高景区的工作效率。 C、差异化经营，降低成本 通过对不同区域的客流量统计，可以使管理人员统计各个区域的吸引率和繁忙度，从而对服务人员进行合理分布，提高销售量，降低成本。 显示当前客流状态和变化趋势，对流量较大的区域采取预防突发事件的措施，并可实时观察景区当前的停留人数，从而对电力、维护人员及安防人员等进行合理调整，控制运做成本。
1.项目需求 随着社会的不断进步和经济建设的迅速发展，人们的安全防范意识在不断加强，视频监控系统也就越来越多的应用于各个行业和方面。在视频监控领域中，数字化、网络化、智能化是一种发展趋势。 现有传统的视频监控系统采用的是被动的解决方案，系统只是负责现场视频采集、信息视频传输和视频显示，硬盘录像机（DVR）主要是对视频信息进行数字化存储和传输实时视频流给远程监控端，都需要通过值班人员实时查看众多视频信息来发现现场问题。人们希望视频监控系统能像眼睛一样时刻保持警戒，但由于视频太多而监控人员有限，且长时间盯着监视画面容易疲劳而导致疏忽，监控人员根本无力监看成百上千路摄像头的视频信息，从而可能漏掉潜在的威胁事件，造成重大事故。而智能视频监控设备就可以解决以上这些人为因素，可以长时间有效的监督视频画面，因此，智能化是视频监控的必然趋势。 智能视频监控是以数字化、网络化视频监控为基础，但又有别于一般的网络化视频监控，它是一种更高端的视频监控应用。智能视频监控系统能够自动识别不同的物体，发现监控画面中的异常情况，并能够以最快和最佳的方式发出警报和提供有用信息，从而能够更加有效的协助安全人员处理危机，并最大限度的降低误报和漏报现象。 看守所智能视频监控系统应满足如下需求： 1)重要场所无视频监控死角；实现智能分析的点位需要摄像机参数、部署 角度和高度符合一定标准。 2)围墙周界：在看守所围墙外侧或内侧一米到三米处，24小时不隔断实时 监控，只要有人进入防区、或者是遗留物品，马上报警； 3)监舍：监舍是监控的重点，24小时不间断实时监控，一旦有发生打架斗 殴事件、在夜间（或者规定时间）随意离开自己的床位（包括上厕所）或聚众，系统即时报警，监控中心管理人员可以给予关注以防止事态不可控制； 4)出入口：24小时不间断实时监控，在没有得到授权或者允许的情况下靠 近出入口或者在出入口附近遗留物品时，系统即时报警，监控中心管理人员可以呼叫相应人员前去处理；同时在进监舍的出入口或者其它相应的地方实现人数统计，以便管理人员及时得知是否有犯人没有返回；
产品介绍 版本：1.0.0 日期：2013.07
版本更新 Version.1.0.0 日期：2013.06 内容：针对我公司产品做出较详细的描述。
目录 版本更新 (2) 1智能视频分析 (5) 1.1行业背景和现实意义 (5) 1.2智能视频分析简介 (5) 2算法功能模块 (6) 2.1基本功能 (6) 2.1.1区域入侵检测 (6) 2.1.2拌线检测功能 (7) 2.1.3物品遗留丢失检测 (7) 2.1.4烟火烟雾检测 (8) 2.1.5逆向运动检测 (9) 2.1.6徘徊检测 (9) 2.1.7速度异常检测 (10) 2.1.8跌倒检测 (10) 2.1.9虚拟围墙检测 (10) 2.1.10人群聚集检测 (11) 2.1.11区域计数 (11) 2.1.12暴力行为检测 (12) 2.1.13人数统计 (13) 2.1.14车流密度统计 (13) 2.1.15停车检测 (14) 2.1.16数车 (15) 2.1.17高速公路行人检测 (15) 2.1.18PTZ跟踪 (16) 2.2摄像头诊断功能 (17) 2.2.1摄像头无信号 (17) 2.2.2摄像头遮挡检测 (17) 2.2.3摄像头移动检测 (18) 2.2.4摄像头模糊检测 (18) 2.2.5摄像头亮度异常 (19) 2.2.6摄像头色度异常 (19) 2.2.7摄像头滚屏和分屏 (19) 2.2.8摄像头画面冻结 (20) 2.2.9摄像头信号剧变检测 (20) 2.2.10摄像头噪声检测 (21) 2.2.11摄像头PTZ运动检测 (21) 2.2.12摄像头抖动检测 (22) 2.2.13摄像头雪花检测 (22) 2.2.14摄像头波纹检测 (23) 2.3特殊功能 (23)
智能视频分析系统使用手册 特别提示 本手册介绍产品为智能视频分析系统。智能视频分析系统系统采用智能值守器。为避免由于安装、调试、使用等人为及环境原因而引起的失误及故障,我们特作以下重要提示,请安装和使用本产品的相关人员务必阅读: 1 在安装使用前,须仔细阅读该产品使用手册。 2 在使用本软件时请确认您电脑的IP 与设备的IP 处于同一网段内, 设备IP 默认为192.168.1.254请将您电脑的IP 设置在192.168.1.*网段。 3 请您连接好设备后改为您需要的IP 地址。为了保证连接正常,请您最好能够使用静态IP 地址。 软件版本2.0.0.3 目录 1. 序言.................................................................................................................................... .. (1) 2. 软件安装.................................................................................................................................... (2) 3. 配置管理工具软件 (7) 3. IVC 智能分析配置工具---使用说明 (7) 一.IVC 概述.................................................................................................................................... . (7) 二.功能简介.................................................................................................................................... (7) 1. 报警系统 (7) 2. 设备控制 (7) 3. 视频通道 (7)