The steps of combining Praat for speech experiment

Recently updated several Praat scripts to extract data from the callout's textgrid some scripts, found that some friends will ask more details of the problem, so there is an idea to combine praat for a phonetics commonly used in the experimental study of the steps are sorted out, hoping for the needs of friends to help. In fact, the most likely to engage in speech research is a major in linguistics, some software, the use of scripts may not be as fast as engineering students, and if it is a technical background if you master this kind of software, script, may disdain to see too careful explanation, this article is aimed at the subject of partial linguistics, and strive to let everyone in the experimental research, It's not too obsessed with how to use it, but it's quicker to complete your experiment. Welcome to ask questions, I will update at any time in order to be able to improve. Here are some of the major operations that I have listed, or the related studies that need to be done, if there are more requirements that could not be listed, please note.

Preparation: Basic use of Praat

Praat has become a more popular voice processing software, it is also very convenient to use, there are many similar tutorials on the internet, the most famous when it is a teacher of the language of the Academy of Social Sciences, teachers can be downloaded to the official language website, do not trust the information of individual websites to buy this tutorial. I'm here to mention a few simple operations, open files, annotate files, recognize several key features of the spectrogram, as well as save the file, other operations such as interested can download the teacher Bear's tutorial, study carefully.

Open File

1. Open the Praat---The open---Read from file ...---find the corresponding sound or textrid file, open it

2. After opening to the Praat window, create a blank callout file

Annotate---to textgrid ...

Note that planning ahead of time you need to label this sound file a few layers of information, usually monogram phonetic symbols, syllables, or word information, prosodic information, or some other information, any number of layers can be set, here only for example set phoneme layer and syllable layer.

3. Select both the sound file and the Textgrid file, click View & Edit to annotate, and when labeling, determine the boundary of the phoneme or syllable according to the boundaries of the hearing and the information of the view. The specific operation can refer to the teacher Bear's tutorial.

4. Understanding the main features of the language map, if you do not see the basic frequency line, Resonance Peak line or rimal line, using the above menu show pitch, show formant, show intersity on it.

5. Save the file

Praat---Save---Saved as text file ...---Saves the callout file as * * *. Textgrid can be.

First step: Recording stage

Phonetics experiment must be inseparable from the recording, there are many options, but this step is not the focus of this article, because each person may request, the conditions are different, theoretically we recommend that all the recordings are high fidelity, that is, in the professional recording studio, to ensure that all the sound in uniform conditions, almost no noise control of the minimum, Such sounds are the best. But not everyone will have such a condition, for example, if a person to do a small dialect of the study, perhaps the area of this dialect is relatively remote areas, these places even travel is more difficult, it may be difficult to have such a professional recording venues, and because of the choice of recording objects, It may not be possible to put all the recordings to the professional recording studio, so this time only to retreat and second, with some equipment to their side, there is a special field called "Field phonetics." So at this point, you can only guarantee your own equipment, as far as possible to achieve better noise resistance. I am not a more professional person, this aspect may be able to consult the journalist category of staff, they have some portable equipment can be used. In addition, you can also use your own notebook, an external sound card, a relatively professional microphone, also enough.

No matter what the way, ultimately we need these devices + software to collect the sound files, usually we need to be WAV format, try not to use the format MP3, which is compressed. WAV format generally note that the sampling rate of more than 16K, the specific recording process please search for more professional articles, if it is a personal computer recording, recommended to use cooledit or Adobe Audition recording.

Also in the recording of a detail, I believe that everyone's research is not a simple few voices, but in batches, such as hundreds of voices, or thousands, each sound unit may be a vowel consonant, a word, a word, a sentence ... This is the way to pay attention to the recording. One way is if you record a unit, such as recording a word "start", then stop recording, and then save the sound for example test001_ start. wav, then the next word, if you record thousands of voices, it is not enough to be recorded! So the general situation, we need to give a list, and then let the recording people have been recorded, and finally the big sound to be sliced, cut into such small units. Why do you want to slice it? Because no segmentation, whether it is labeled, or extract data, etc. are very inconvenient, and is not conducive to retrieval.

Here involves a segmentation method problem, you can again through the stupid method, with Cooledit Open this long sound, a voice of a voice of choice, and then save for their own needs of the name, thousands of files down, believe that has been dizzy, and prone to error. The next step is to recommend a relatively dexterous approach, which is of course optional. In addition to the premise of this method you have to make sure that there are enough pauses between each unit in the recording, and not connected together. Here is an example of how it is easier to use auto-segmentation in this way.

Step Two: Long sound segmentation

Please refer to the following blog post: Praat to cut a continuously recorded sound file into small unit files

http://blog.csdn.net/shaopengfei/article/details/20928683

Step three: Manually annotate the Voice

Now with the sound, is to be labeled, first introduced a full manual labeling, this method may be troublesome, the cost of time, but may be accurate rate is self-control, the following fourth step to introduce the automatic labeling, although the generation, but the accuracy rate must also be manual intervention, The advantage is that there is no need to manually add a lot of boundaries, so I still recommend the use of automatic labeling methods, and then manually carefully adjust the boundaries. One way to do this is to open the Praat, then open a sound, create a blank callout file, set a few layers according to your needs, and then label the information separately. Then save the callout file. This can be tedious, and a lot of time is spent on generating blank textgrid, generating boundaries, and saving operations. I have provided a tool that automatically generates textgrid through this tool, then generates an empty phoneme boundary, adjusts well, and then saves it.

Please refer to the following blog post: Tools for labeling with auxiliary Praat

http://blog.csdn.net/shaopengfei/article/details/43020707

Fourth step: Automatically annotate speech Sppas

The second step of the cut out of the sound file, the same will generate a recorded text, so that the Sppas tool can be used to automatically generate the original callout file.

Please refer to the following blog post: Voice callout auto Segment Alignment tool Sppas using notes

http://blog.csdn.net/shaopengfei/article/details/18351809

Fifth step: Manually adjust the results of automatic labeling

This step refers to the third step.

Sixth step: Artificial repair of the base frequency

Seventh step: Extracting Parameters-base frequency

Eighth step: Extracting parameters-resonant peaks

Nineth Step: Acoustic vowel image

Tenth step: Sentence Intonation chart

11th Step: Chinese Character tone chart

[Update ...]

The steps of combining Praat for speech experiment