Giters

tjmahr / mfa-dict

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

mfa-dict

This repository hosts the words we have had to add to the English pronunciation dictionary for the Montreal Forced Aligner (Docs, GitHub.

contents

Here are the contents of the custom dictionary, as an R data-frame, where I have named unimportant columns "ignoreX".

"english_us_arpa_extras.dict" |> 
  readr::read_tsv(
    col_names = c("word", "prob", "ignore0", "ignore1", "ignore2", "phones"),
    col_types = "cddddc"
  ) |> 
  print(n = Inf)
#> # A tibble: 6 × 6
#>   word     prob ignore0 ignore1 ignore2 phones            
#>   <chr>   <dbl>   <dbl>   <dbl>   <dbl> <chr>             
#> 1 hotdog   0.99    0.17       1       1 HH AA1 T D AO2 G  
#> 2 hotdogs  0.99    0.17       1       1 HH AA1 T D AO2 G Z
#> 3 standed  0.99    0.17       1       1 S T AE1 N D IH0 D 
#> 4 maked    0.99    0.17       1       1 M EY1 K T         
#> 5 dranked  0.99    0.17       1       1 D R AE1 NG K T    
#> 6 baggie   0.99    0.17       1       1 B AE1 G IY0

To use the dictionary, copy and paste the contents of file english_us_arpa_extras.dict onto the end of the dictionary file used by the MFA.

updating/modifying the dictionary

I used the following to force an update to my English models.

mfa model download dictionary english_us_arpa --ignore_cache
mfa model download acoustic english_us_arpa --ignore_cache

So, for my Windows computer, the dictionary was saved to:

C:/Users/Tristan/Documents/MFA/pretrained_models/dictionary/english_us_arpa.dict

Then in R I can read in the MFA dictionary, our custom one, and combine them and overwrite the MFA one.

# Using the convention that ~ points to a user's `Documents` folder on Windows
path_mfa_dict <- "~/MFA/pretrained_models/dictionary/english_us_arpa.dict"

current <- readr::read_tsv(
  path_mfa_dict,
  col_names = FALSE,
  col_types = "cddddc"
)

new <- readr::read_tsv(
  "english_us_arpa_extras.dict",
  col_names = FALSE,
  col_types = "cddddc"
)

# Add or overwrite rows in `current` with rows from `new`
dplyr::rows_upsert(current, new, by = c("X1", "X6")) |> 
  readr::write_tsv(path_mfa_dict, col_names = FALSE)

the dictionary format

The format for the probabilistic pronunciation dictionaries is [described here][mfa-pron-prob]. Consider the word the:

the 0.99    0.01    1.21    0.97    DH AH0
the 0.01    0.12    1.87    0.84    DH AH1
the 0.17    0.11    1.27    0.96    DH IY0

The docs tell us:

The first float column is the probability of the pronunciation, the next float is the probability of silence following the pronunciation, and the final two floats are correction terms for preceding silence and non-silence. Given that each entry in a dictionary is independent and there is no way to encode information about the preceding context, the correction terms are calculated as how much more common was silence or non-silence compared to what we would expect factoring out the likelihood of silence from the previous word.

I am not sure what this means, but the first column seems to be the most important (marginal) probability of pronunciation.

About