ps-vs-hmong-female.Rmd

---
title: 'PraatSauce/VoiceSauce comparison: White Hmong (female)'
author: "James Kirby"
date: "31/01/2018"
output:
  pdf_document:
    toc: yes
    toc_depth: '3'
  html_document:
    toc: yes
    toc_depth: 3
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, cache = TRUE)
require(ggplot2)
require(plyr)
require(tidyr)
require(stringr)
library(scales)
theme_set(theme_bw())
```

```{r include=FALSE}

######################
## VoiceSauce 
######################

## VoiceSauce columns we care about
keeps <- c('Filename', 'Label', 'Item', 'seg_Start', 'seg_End', 't_ms', 'H1c', 'H2c', 'H4c', 'A1c', 'A2c', 'A3c', 'H1u', 'H2u', 'H4u', 'A1u', 'A2u', 'A3u', 'H2Ku', 'H5Ku', 'CPP', 'HNR05', 'HNR15', 'HNR25', 'HNR35', 'H1H2u', 'H2H4u', 'H1A1u', 'H1A2u', 'H1A3u', 'H1H2c', 'H2H4c', 'H1A1c', 'H1A2c', 'H1A3c', 'H2KH5Ku', 'strF0', 'sF0', 'pF0', 'sF1', 'sF2', 'sF3', 'pF1', 'pF2', 'pF3', 'pB1', 'pB2', 'pB3', 'sB1', 'sB2', 'sB3')

## VoiceSauce, formula bandwidths
vs.fbw <- read.csv('/Users/jkirby/Documents/Projects/praatsauce/comp/hmong_female_voicesauce_formulabw.txt', sep='\t')
## get rid of .mat extention
vs.fbw$Filename <- gsub('.mat', '', vs.fbw$Filename)
vs.fbw <- vs.fbw %>% separate(Filename, c("Num", "Item", "Junk"), "-", remove=FALSE)
vs.fbw <- vs.fbw[keeps]

## scaled time
vs.fbw<-ddply(vs.fbw, .(Filename), mutate, t=rescale(t_ms))

vs.fbw$method = rep("formula")
vs.fbw$script = rep("VoiceSauce")
vs.fbw.long <- gather(vs.fbw, measure, value, H1c:sB3, factor_key=TRUE)        

## VoiceSauce, estimated bandwidths (maybe)
vs.ebw <- read.csv('/Users/jkirby/Documents/Projects/praatsauce/comp/hmong_female_voicesauce_estbw.txt', sep='\t')
## get rid of .mat extention
vs.ebw$Filename <- gsub('.mat', '', vs.ebw$Filename)
vs.ebw <- vs.ebw %>% separate(Filename, c("Num", "Item", "Junk"), "-", remove=FALSE)
vs.ebw <- vs.ebw[keeps]

## scaled time
vs.ebw<-ddply(vs.ebw, .(Filename), mutate, t=rescale(t_ms))

vs.ebw$method = rep("estimated")
vs.ebw$script = rep("VoiceSauce")
vs.ebw.long <- gather(vs.ebw, measure, value, H1c:sB3, factor_key=TRUE)        


######################
## PraatSauce
######################

## symmetric kernel
f21 <- rep(1/21,21)
## lag kernel
f20 <- rep(1/20,20)

## formula bandwidths
ps.fbw <- read.csv('/Users/jkirby/Documents/Projects/praatsauce/comp/hmong_female_praatsauce_formulabw.txt', header=TRUE)
## turn into msec
ps.fbw$seg_Start <- ps.fbw$seg_Start * 1000
ps.fbw$seg_End <- ps.fbw$seg_End* 1000
ps.fbw$t_ms <- ps.fbw$t_ms* 1000
drops <- c('t', 'var1', 'var3')
ps.fbw <- ps.fbw[, !(names(ps.fbw) %in% drops)]

names(ps.fbw) <- c('Filename', 'Item', 'Label', 'seg_Start', 'seg_End', 't_ms', 'pF0', 'pF1', 'pF2', 'pF3', 'pB1', 'pB2', 'pB3', 'H1u', 'H2u', 'H4u', 'H2Ku', 'H5Ku', 'A1u','A2u', 'A3u', 'H1H2u', 'H2H4u', 'H1A1u', 'H1A2u', 'H1A3u', 'H2KH5Ku', 'H1c', 'H2c', 'H4c', 'A1c', 'A2c', 'A3c', 'H1H2c', 'H2H4c', 'H1A1c', 'H1A2c', 'H1A3c', 'CPP', 'HNR05', 'HNR15', 'HNR25',  'HNR35')

## remove zeros
ps.fbw[ps.fbw == 0] <- NA

## smoothing
#ps.fbw <- cbind(ps.fbw[1:6], apply(ps.fbw[7:43], 2, filter, filter=f21, sides=2))

## scaled time
ps.fbw<-ddply(ps.fbw, .(Filename), mutate, t=rescale(t_ms))

ps.fbw$method = rep("formula")
ps.fbw$script = rep("PraatSauce")
ps.fbw.long <- gather(ps.fbw, measure, value, pF0:HNR35, factor_key=TRUE)

## estimated bandwidths
ps.ebw <- read.csv('/Users/jkirby/Documents/Projects/praatsauce/comp/hmong_female_praatsauce_estbw.txt', header=TRUE)
## turn into msec
ps.ebw$seg_Start <- ps.ebw$seg_Start * 1000
ps.ebw$seg_End <- ps.ebw$seg_End* 1000
ps.ebw$t_ms <- ps.ebw$t_ms* 1000
ps.ebw <- ps.ebw[, !(names(ps.ebw) %in% drops)]

names(ps.ebw) <- c('Filename', 'Item', 'Label', 'seg_Start', 'seg_End', 't_ms', 'pF0', 'pF1', 'pF2', 'pF3', 'pB1', 'pB2', 'pB3', 'H1u', 'H2u', 'H4u', 'H2Ku', 'H5Ku', 'A1u','A2u', 'A3u', 'H1H2u', 'H2H4u', 'H1A1u', 'H1A2u', 'H1A3u', 'H2KH5Ku', 'H1c', 'H2c', 'H4c', 'A1c', 'A2c', 'A3c', 'H1H2c', 'H2H4c', 'H1A1c', 'H1A2c', 'H1A3c', 'CPP', 'HNR05', 'HNR15', 'HNR25',  'HNR35')

## remove zero
ps.ebw[ps.ebw == 0] <- NA

## smooth
#ps.ebw <- cbind(ps.ebw[1:6], apply(ps.ebw[7:43], 2, filter, filter=f21, sides=2))

## scale
ps.ebw<-ddply(ps.ebw, .(Filename), mutate, t=rescale(t_ms))

ps.ebw$method = rep("estimated")
ps.ebw$script = rep("PraatSauce")

ps.ebw.long <- gather(ps.ebw, measure, value, pF0:HNR35, factor_key=TRUE)

## combine
df <- rbind(ps.fbw.long,ps.ebw.long,vs.fbw.long,vs.ebw.long)

df$corrected <- rep('uncorrected')
df[df$measure %in% c('H1c', 'H2c', 'H4c', 'A1c', 'A2c', 'A3c', 'H1H2c', 'H2H4c', 'H1A1c', 'H1A2c', 'H1A3c'),]$corrected <- 'corrected'
```

## Data

This small dataset compares spectral measures generated by both PraatSauce v0.2.2 and VoiceSauce v1.31 at 1 msec intervals for 9 White Hmong lexical items spoken by a single female speaker. The original audio files can be found [here](http://www.phonetics.ucla.edu/voiceproject/voice.html). For both scripts, 5 formants were estimated with a maximum formant frequency of 5500 Hz; minimum and maximum F0 values were set to 50 Hz and 600 Hz for all F0 estimators. For VoiceSauce, the STRAIGHT F0 estimate and Snack formant/bandwidth estimates were used for harmonic amplitude corrections.

The `method` column indicates whether the formant bandwidths were `estimated` using Praat (PraatSauce) or Snack (VoiceSauce), or whether the Hawks and Miller `formula` was used.

In Hmong orthography, final *-g* indicates a low-falling breathy tone, while *-m* indicates creaky tone.

```{r}
head(df)
```

In the plots which follow, the PraatSauce measures are unsmoothed. If you want to compare to smoothed estimates, uncomment the two lines:

```{r, execture=FALSE}
ps.fbw <- cbind(ps.fbw[1:6], apply(ps.fbw[7:43], 2, filter, filter=f21, sides=2))
ps.ebw <- cbind(ps.ebw[1:6], apply(ps.ebw[7:43], 2, filter, filter=f21, sides=2))
```

This implements a symmetric kernel filter. This is different from what VoiceSauce does. VoiceSauce uses the Matlab `filter()` function, by default a lag filter which pads with zeros. So while the smoothed value of sample 20 is equal to $\sum_{i=1}^{20}/20$, the smoothed value of sample 19 is not undefined, but is calculated as $\sum_{i=1}^{19}/20$, the smoothed value of sample 18 is $\sum_{i=1}^{18}/20$, etc.

If you want to smooth the Matlab way, use the lag kernel by selecting `filter=f20` and set `sides=1`.

## Plots

### F0

```{r, echo=FALSE}
ggplot(subset(df, value > 0 & method == 'formula' & measure %in% c('strF0', 'pF0', 'sF0')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free') 
```

All F0 estimators except for STRAIGHT have difficulty with the somewhat constricted vowel quality of *cav* 'to argue'. 

\newpage
### Formants

```{r, echo=FALSE}
ggplot(subset(df, method == 'estimated' & measure %in% c('pF1', 'pF2', 'pF3')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free') + ggtitle("Formants estimate using Praat")

ggplot(subset(df, (script=='PraatSauce' & method == 'estimated' & measure %in% c('pF1', 'pF2', 'pF3')) | (script=='VoiceSauce' & method == 'estimated' & measure %in% c('sF1', 'sF2', 'sF3'))), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure, script))) + geom_line() + facet_wrap(~Filename, scales='free') + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3)))) + ggtitle("Formants estimated using Snack (VoiceSauce) vs. Praat (PraatSauce)")
```

\newpage
### Bandwidths

#### PraatSauce estimated vs. formula bandwidths 

```{r, echo=FALSE}
ggplot(subset(df, script=='PraatSauce' & measure %in% c('pB1', 'pB2', 'pB3')), aes(t_ms, value, colour=measure, linetype=method, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free')
```

Compared to the formula estimates, PraatSauce estimated bandwidths are huge...

\newpage
#### PraatSauce vs. VoiceSauce estimated bandwidths

```{r, echo=FALSE}

ggplot(subset(df, method == 'estimated' & measure %in% c('pB1', 'pB2', 'pB3')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free')
```

... but VoiceSauce Praat-estimated bandwidths are an order of magnitude huger.

\newpage
#### VoiceSauce Praat vs. Snack estimated bandwidths

```{r, echo=FALSE}
ggplot(subset(df, method == 'estimated' & script=='VoiceSauce' & measure %in% c('pB1', 'pB2', 'pB3', 'sB1', 'sB2', 'sB3')), aes(t_ms, value, colour=measure, linetype=measure, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free') + scale_linetype_manual(values=c('solid', 'solid', 'solid', 'dashed', 'dashed', 'dashed')) + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3))))
```

VoiceSauce's Snack estimates (if that's really what they are) look less erratic.

\newpage
####  VoiceSauce Snack vs. PraatSauce estimated bandwidths 

```{r, echo=FALSE}
ggplot(subset(df, method == 'estimated' & (script=='PraatSauce' & measure %in% c('pB1', 'pB2', 'pB3')) | (script == 'VoiceSauce' & measure %in% c('sB1', 'sB2', 'sB3'))), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free') + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3))))
```

PraatSauce estimates not completely off from Snack's.

\newpage
### Uncorrected amplitudes

#### PraatSauce vs. VoiceSauce H1, H2, H4
Note that the choice of bandwidth estimator is irrelevant here.

```{r, echo=FALSE}
ggplot(subset(df, method == 'formula' & measure %in% c('H1u', 'H2u', 'H4u')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x")
```

The middle third of *cav* is a real problem for PraatSauce (at least with the chosen settings).

\newpage
#### PraatSauce vs. VoiceSauce A1, A2, A3

```{r, echo=FALSE}
ggplot(subset(df, method == 'formula' & measure %in% c('A1u', 'A2u', 'A3u')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free_x') 
```

The higher-order harmonics are not as much of a problem.

VoiceSauce estimates are consistently 20-25 dB lower than the PraatSauce estimates, and are sometimes negative, which seems...strange. This suggests to me they are being attenuated somewhere, though I have not been able to find the piece of code where this happens.

\newpage
### Corrected amplitudes

Here, choice of formant bandwidth estimator potentially matters.

In these plots, PraatSauce is using Praat and VoiceSauce is using Snack estimates.

#### PraatSauce vs. VoiceSauce H1\*, H2\*, H4\*

```{r, echo=FALSE}
ggplot(subset(df, method=='estimated' & measure %in% c('H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") + ggtitle("Estimated bandwidths")

ggplot(subset(df, method == 'formula' & measure %in% c('H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") +ggtitle("Formula bandwidths")
```

For VoiceSauce, using estimated bandwidths is virtually unnoticeable:

\newpage
#### VoiceSauce estimated vs. formula bandwidths, H1\*, H2\*, H4\*

```{r, echo=FALSE}
ggplot(subset(df, script=='VoiceSauce' & measure %in% c('H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=method, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") 
```

For PraatSauce, using the formula bandwidths makes only very minor differences:

\newpage
#### PraatSauce estimated vs. formula bandwidths, H1\*, H2\*, H4\*

```{r, echo=FALSE}
ggplot(subset(df, script=='PraatSauce' & measure %in% c('H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=method, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") 
```

\newpage
#### PraatSauce vs. VoiceSauce A1\*, A2\*, A3\*

```{r, echo=FALSE}
ggplot(subset(df, method=='estimated' & measure %in% c('A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") + ggtitle("Estimated bandwidths")
```

\newpage
#### PraatSauce vs. VoiceSauce A1\*, A2\*, A3\*
```{r, echo=FALSE}
ggplot(subset(df, method == 'formula' & measure %in% c('A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") + ggtitle("Formula bandwidths")
```

\newpage
#### VoiceSauce estimated vs. formula bandwidths, A1\*, A2\*, A3\*

```{r, echo=FALSE}
ggplot(subset(df, script=='VoiceSauce' & measure %in% c('A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=method, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") 
```

\newpage
#### PraatSauce estimated vs. formula bandwidths, A1\*, A2\*, A3\*

```{r, echo=FALSE}
ggplot(subset(df, script=='PraatSauce' & measure %in% c('A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=method, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") 
```

\newpage
#### PraatSauce corrected vs. uncorrected

```{r, echo=FALSE}


ggplot(subset(df, script=='PraatSauce' & method=='estimated' & measure %in% c('H1u', 'H2u', 'H4u', 'H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=corrected, group=interaction(corrected,method,measure))) + geom_line() + facet_wrap(~Filename, scales="free_x") + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3)))) + ggtitle("Estimated bandwidths")

ggplot(subset(df, script=='PraatSauce' & method=='estimated' & measure %in% c('A1u', 'A2u', 'A3u', 'A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=corrected, group=interaction(corrected,method,measure))) + geom_line() + facet_wrap(~Filename, scales="free_x") + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3)))) + ggtitle("Formula bandwidths")
```

\newpage
#### VoiceSauce corrected vs. uncorrected

```{r, echo=FALSE}
ggplot(subset(df, script=='VoiceSauce' & method=='estimated' & measure %in% c('H1u', 'H2u', 'H4u', 'H1c', 'H2c', 'H4c')), aes(t_ms, value, colour=measure, linetype=corrected, group=interaction(corrected,method,measure))) + geom_line() + facet_wrap(~Filename, scales="free_x") + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3)))) + ggtitle("Estimated bandwidths")

ggplot(subset(df, script=='VoiceSauce' & method=='estimated' & measure %in% c('A1u', 'A2u', 'A3u', 'A1c', 'A2c', 'A3c')), aes(t_ms, value, colour=measure, linetype=corrected, group=interaction(corrected,method,measure))) + geom_line() + facet_wrap(~Filename, scales="free_x") + scale_color_manual(values=c(c(hue_pal()(3)), c(hue_pal()(3)))) + ggtitle("Formula bandwidths")
```

\newpage
### Corrected differences

#### PraatSauce vs. VoiceSauce H1\*-H2\* & H2\*-H4\*

```{r, echo=FALSE}
ggplot(subset(df, method == 'estimated' & measure %in% c('H1H2c', 'H2H4c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") +ggtitle("Estimated bandwidths")

ggplot(subset(df, method == 'formula' & measure %in% c('H1H2c', 'H2H4c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales="free_x") + ggtitle("Formula bandwidths")
```

\newpage
##### PraatSauce vs. VoiceSauce H1\*-A1\*, A2\*, A3\*

```{r, echo=FALSE}
ggplot(subset(df, method=='estimated' & measure %in% c('H1A1c', 'H1A2c', 'H1A3c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free_x') + ggtitle("Estimated bandwidths")

ggplot(subset(df, method=='formula' & measure %in% c('H1A1c', 'H1A2c', 'H1A3c')), aes(t_ms, value, colour=measure, linetype=script, group=interaction(method,measure,script))) + geom_line() + facet_wrap(~Filename, scales='free_x') + ggtitle("Formula bandwidths")
```

PraatSauce seems to have higher difference estimates for (some of) the *-g* items.

The issue with the middle third of *cav* might be regarded as positive if this token is really being produced with nonmodal voice.

\newpage

### Cepstral peak prominence

```{r, echo=FALSE}
ggplot(subset(df, method=='formula' & measure=='CPP'), aes(t_ms, value, group=script, colour=script)) + geom_line() + facet_wrap(~Filename, scales='free_x')
```

Praat(Sauce) estimates are comparable if smoothed.

\newpage
### Harmonic to noise ratios

Here just showing HNR05 and HNR15 for clarity.

```{r, echo=FALSE}
ggplot(subset(df, value>-10 & method=='formula' & measure %in% c('HNR05', 'HNR15')), aes(t_ms, value, group=interaction(measure,script), colour=measure, linetype=script)) + geom_line() + facet_wrap(~Filename, scales='free_x')
```

Again, the Praat estimates differ in amplitude, but maintain roughly the same trajectories. However, the PraatSauce implementation is much less sophisticated than that of VoiceSauce, and relies entirely on Praat's `To Harmonicity...` function.

\newpage
## Distinguishing voice qualities

```{r, include=FALSE}
is <- subset(df, Item %in% c('tis', 'tig', 'tim'))
as <- subset(df, Item %in% c('cab', 'cag', 'cav'))
os <- subset(df, Item %in% c('pob', 'pog', 'poj'))
```

### High vowels
```{r, echo=FALSE}
ggplot(subset(is, measure %in% c('H1H2c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-H2*")
```
PraatSauce does not really find the expected positive spectral slope for `tig`. It is clearly distinguished by VoiceSauce, but the slope is unexpectedly negative, rather than positive.

```{r, echo=FALSE}
ggplot(subset(is, measure %in% c('H1A1c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-A1*")

ggplot(subset(is, measure %in% c('CPP')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce CPP")
```

### Mid vowels
```{r, echo=FALSE}
ggplot(subset(os, measure %in% c('H1H2c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-H2*")
ggplot(subset(os, measure %in% c('H1A1c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-A1*")
ggplot(subset(os, measure %in% c('CPP')), aes(t, value, colour=Item, linetype=method, group=interaction(Item,measure,method))) + geom_line() + facet_grid(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce CPP")
```

For the mid vowel, the breathy toned `pog` is more clearly differentiated, though again VoiceSauce makes a clearer distinction.

### Low vowels
```{r, echo=FALSE}
ggplot(subset(as, measure %in% c('H1H2c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-H2*")
ggplot(subset(as, measure %in% c('H1A1c')), aes(t, value, colour=Item, linetype=method,  group=interaction(Item,measure,method))) + geom_line() + facet_wrap(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce H1*-A1*")
ggplot(subset(as, measure %in% c('CPP')), aes(t, value, colour=Item, linetype=method, group=interaction(Item,measure,method))) + geom_line() + facet_grid(~script, scales="free_x") + ggtitle("VoiceSauce vs. PraatSauce CPP")
```

For the low vowel tokens, VoiceSauce doesn't seem to distinguish `cag` from other syllables except for via CPP.