6  Configuring emuR for ToBI annotations

6.1 Objective and preliminaries

The objective of this chapter is to show how to go from a collection of .wav and .txt files to an Emu database that is useful for tones-and-break indices (ToBI) annotation. More specifically, the aim is to configure a database to have a configuration like that shown in Figure 6.1.

Figure 6.1: An Emu database with ToBI annotations.

The starting point is the text collection of Albanian data analyzed in Chapter 5.

We will also see how an annotation level with syllable structures can be ascertained from a database by creating an CV-style annotation level.

The assumption is that you already have an R project called ipsR and that it contains the directories emu_databases and testsample. If this is not the case, please go back and follow the preliminaries chapter.

library(tidyverse)
library(emuR)
library(wrassp)
sourceDir <- "./testsample"
targetDir <- "./emu_databases"

6.2 Forced alignment

The first task is to apply forced alignment to these Albanian data. The commands will be presented in this section without detailed comment given that these were discussed in Chapter 5. The resulting Emu database will be stored as alb2_DB (to distinguish it from the one created in the Chapter 5).

We first create and load the Emu database using convert_txtCollection() and load_emuDB(), and subsequently use runBASwebservice_all() to run the MAUS pipeline over the data.

path.albanian <- file.path(sourceDir, "albanian")
convert_txtCollection(dbName = "alb2", 
                      sourceDir = path.albanian,
                      targetDir = targetDir,
                      verbose = FALSE)
alb2_DB <- load_emuDB(file.path(targetDir, "alb2_emuDB"),
                      verbose=FALSE)
runBASwebservice_all(alb2_DB,
                     transcriptionAttributeDefinitionName = "transcription", 
                     language = "sqi-AL",  
                     runMINNI = FALSE,
                     verbose = FALSE)

6.3 Configuring a database for ToBI annotations

The database currently has an ITEM annotation level ORT that dominates an ITEM level MAS containing syllabifications. MAS dominates the SEGMENT level MAU, which contains the phonetic segmentation. You can confirm this either by serve()ing the database or by calling summary(alb2_DB).

Our next task is to add a level Tone for marking intonational events and to add a new SEGMENT level, ORT2 which contains word-level segmentations. The tonal events of the Tone tier and annotations of the ORT2 tier are to be linked automatically, just as in Figure 6.1.

Here are the steps needed:

  • Create the SEGMENT level ORT2 containing word annotations linked to times.
  • Create a Tone level and link it to ORT2.
  • Calculate and display pitch tracks.
  • Annotate the Tone level and link the annotations automatically to the word level annotations in ORT2.

6.3.1 Create and display word level annotations

This is a four step process:

  • Create a SEGMENT-type annotation level called ORT2.

As covered in Chapter 4, this can be done with the add_levelDefinition() function like so:

add_levelDefinition(alb2_DB, 
                    name="ORT2",
                    type="SEGMENT",
                    verbose=FALSE)

We can confirm that this step was successful with list_levelDefinitions():

list_levelDefinitions(alb2_DB)
  • Make a list of all words in the database.

We can do this by using the query() function to find all annotations in the ORT level.

text.s <- query(alb2_DB, "[ORT =~ .*]")
text.s

We can use the segment list generated by the query() above to populate our new level ORT2. We simply need to change the columns attribute and level so they show the value ORT2 instead of ORT:

text.s$attribute <- "ORT2"
text.s$level <- "ORT2"
  • Add word-level annotations to ORT2.

We can now pass text.s on to the function create_itemsInLevel() like so:

create_itemsInLevel(alb2_DB, text.s,
                    verbose=FALSE)

It is now possible to query() the word level annotations directly from ORT2:

query(alb2_DB, "[ORT2 =~ .*]")
  • Display the annotations of ORT2 time-aligned to the signals.

Because ORT2 is a SEGMENT-type level and not an ITEM-type level, we can now display word-level annotations underneath the signals in the EMU-webApp. However, if we serve() the database now, only the segment-level annotations in MAU can be seen. As covered in Chapter 4, this can be controlled with the set_levelCanvasesOrder() function like so:

set_levelCanvasesOrder(alb2_DB,
                       perspectiveName = "default",
                       order = "ORT2")

serve()ing the database now will show only the word-level annotations in ORT2 underneath the signals:

serve(alb2_DB, useViewer = FALSE)

6.3.3 Calculating and displaying F0

This is a three-step process. Most of these steps were covered in Chapter 4, so only a brief summary is given below.

  • Calculate F0.

We can do this using the add_ssffTrackDefinition with the “on the fly function” mhsF0 (see Chapter 4).

add_ssffTrackDefinition(alb2_DB, 
                        name = "pitch", 
                        onTheFlyFunctionName = "mhsF0",
                        verbose = FALSE)

We can confirm that this worked with the function list_ssffTrackDefinitions():

list_ssffTrackDefinitions(alb2_DB)
  • Configure the database to show the F0 track.

As explained in Chapter 4, the pitch data can be displayed with the function set_signalCanvasesOrder. Currently, only the waveform and spectrogram are displayed, as confirmed with:

get_signalCanvasesOrder(alb2_DB, perspectiveName = "default")
[1] "OSCI" "SPEC"

In order to display the pitch data underneath the spectrogram, use set_signalCanvasesOrder() like follows:

set_signalCanvasesOrder(alb2_DB, 
                        perspectiveName = "default",
                        order = c("OSCI", "SPEC", "pitch"))

(Alternatively, if you have installed and loaded emuhelpeR, you can display it like so: alb2DB %>% add_signal_canvas("pitch").)

Overlaying pitch on the spectrogram

If you have the library emuhelpeR installed, you can overlay pitch on the spectrogram using the signal_on_spec() function like so:

library(emuhelpeR)
alb2_DB %>% signal_on_spec("pitch")

(The function clear_spec() from emuhelpeR can be used to remove the signal from the spectrogram if you no longer wish to display it there.)

If you do not have emuhelpeR installed, the process is a bit more intricate. For this you will usually use a plain text editor to change the database’s configuration JSON file. This file always has the same name as the database + config.json and is stored in the same directory as the database. Once you’ve found this file, you can proceed as follows:

  • Optionally make a backup copy of the the alb2_DBconfig.json file in case anything goes wrong.
  • Open alb2_DBconfig.json with a plain text editor, e.g. Notepad in Windows.
  • Search the file for the text string assign. You should find a line with the following contents:

"assign": [],

  • Carefully replace this line with the following:

"assign": [{ "signalCanvasName": "SPEC", "ssffTrackName": "pitch" }],

  • Save the file.
  • serve() the database again.

In principle, the configuration JSON file can also be edited in R, e.g. using the rjson library. You should probably avoid this unless you’re an advanced R user – it’s a finnicky process and a lot can go wrong.

6.3.4 Automatically linking annotations at the Tone level

The first task here is to make a few annotations. More specifically, add two pitch targets (e.g., L* to Lena, L* to leu) to the utterance 0001BF_1syll_1 as in Figure 6.1. See Chapter 3 for details of how to annotate in the Emu-webApp.

The next task is to automatically link the annotations at the Tone level to the corresponding words in the ORT2 level. We can do this using the function autobuild_linkFromTimes(). This function links an annotation Tone level to an annotation at the ORT2 level if the Tone annotation falls within the segment segment boundaries of the ORT2 annotation. In other words, if the start and end times of an ORT2 annotation are denoted t1 and t2, and the time of a Tone annotation is denoted simply t, then annotations at the two levels are linked if t1 < t < t2.

autobuild_linkFromTimes(alb2_DB,
                        superlevelName = "ORT2",
                        sublevelName = "Tone",
                        verbose = FALSE)

If you serve() the database again and look at the hierarchy view, you can confirm that the links were indeed created.

It should now be possible to query tones with respect to words and vice-versa. For example, this function finds all words at the ORT2 level with an annotated pitch accent at the Tone level. (Much more detail on querying follows in Chapter 7!)

query(alb2_DB, "[Tone =~.* ^ #ORT2 =~.*]")

6.4 Adding a CV level

It can be helpful to have another annotation level in which every vowel corresponds to V and every consonant to C. This could then be used to establish the syllable types in the database. To do this, a new level called CV will be created that is an attribute of the MAU tier (which contains the segment annotations). We do this with the add_attributeDefinition() function:

add_attributeDefinition(alb2_DB,
                        levelName = "MAU",
                        name = "CV",
                        verbose=FALSE)

Here we query all segments found at the MAU level in the database:

mau.s <- query(alb2_DB, "MAU =~ .*")
mau.s

We can find unique segments (and their frequencies) using the count function available in the tidyverse:

count(mau.s, labels)

We’ll store the vowels in a vector vowel_phonemes:

vowel_phonemes = c("a", "E", "i", "O", "u", "4")

With a few tidyverse commands, we can wrangle mau.s to have the label V for vowels and C for consonants (i.e., the labels that are not in the vowel_phonemes vector), and storing this as a new data frame called new_items.

We first change the values in the attribute column to CV using the mutate() function. We use the rename() function to change the labels column with phonemic annotations to old_labels. Finally we use mutate() in combination with case_when() to store the label V in labels wherever old_labels has a phonemic annotation found in our vowel_phonemes vector, and to store the label C in all other cases. For more resources on the tidyverse see the initial setup chapter.

new_items <- mau.s %>%
  mutate(attribute = "CV") %>%
  rename(old_labels = labels) %>%
  mutate(labels = case_when(old_labels %in% vowel_phonemes ~ "V",
                            TRUE ~ "C"))

Calling the count() function on new_items will give us the number of consonants and vowels in the database:

count(new_items, labels)

We can add these new annotations to the attribute level CV using the update_itemsInLevel() function like so:

update_itemsInLevel(alb2_DB, 
                    itemsToUpdate = new_items,
                    verbose = FALSE)

We can now query() the CV level for all of its annotations:

seglist_cv <- query(alb2_DB, "CV =~ .*")
count(seglist_cv, labels)

Perhaps more importantly, it is now possible to identify the different types of syllable structure. Recall that the MAS tier parses the Phoneme tier into syllables.

syll <- query(alb2_DB, "MAS =~ .*")
count(syll, labels)

Here are the syllables in terms of their representation at the CV tier:

syll.cv <- requery_hier(alb2_DB, syll, level = "CV")
count(syll.cv, labels)

Don’t worry if you don’t yet understand how the querying functions work – much more on that in Chapter 7!

6.5 Functions introduced in this chapter

  • signal_on_spec(): Overlays a signal, such as pitch, on top of the spectrogram in the Emu-webApp. (Requires emuhelpeR to be installed).
  • count(): Provides a count of all instances of a factor variable in a data frame or tibble.
  • requery_hier(): Requeries an emuDB based on an existing query. Will be discussed much more in Chapter 7.