Efficient log-likelihood calculation in depmixS4 package

Question

I want to calculate log likelihood for my model but don't know how to do it efficiently. Here is my code that does what I need, but it's slow

up <- rnorm(300,mean = 1,sd = 1)
dw <- rnorm(300,mean = -1,sd = 1)
test <- c(up[1:100],dw[1:100],up[101:200],dw[101:200],up[201:300],dw[201:300])

train <- data.frame(y=up)
test <- data.frame(y=test)

library(depmixS4)
hmm <- depmix(y~1, data=train, nstates=2, family=gaussian()) |> fit()

The function for calculating logLike is extremely slow because it needs to initialize the model with new data every time

logLike <- function(model, new_data){
  init <- depmix(y~1, data=new_data, nstates=2, family=gaussian())
  sp <- setpars(init , getpars(model))
  forwardbackward(sp)$logLike
}

llh <- rep(NA,nrow(test))
for(i in 20:nrow(test)){
  data <- data.frame(y=test[  (i-19):i  ,])
  llh[i] <- logLike(hmm, data)
}

layout(1:2)
plot(cumsum(test$y),t="l")
plot(llh,t="l")

Is there any other way to calculate the logLike more efficiently