Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/tjmahr/mfa-dict


https://github.com/tjmahr/mfa-dict

Last synced: about 1 month ago
JSON representation

Awesome Lists containing this project

README 

        
---

output: github_document

---



```{r}

#| include: false

library(tidyverse)

knitr::opts_chunk$set(

  collapse = TRUE,

  comment = "#>",

  dpi = 300, 

  dev = "ragg_png"

)

```



# mfa-dict



This repository hosts the words we have had to add to the English

pronunciation dictionary for the Montreal Forced Aligner

([Docs][mfa-docs], [GitHub][mfa-github].



## contents



Here are the contents of the custom dictionary, as an R data-frame,

where I have named unimportant columns `"ignoreX"`.



```{r}

"english_us_arpa_extras.dict" |> 

  readr::read_tsv(

    col_names = c("word", "prob", "ignore0", "ignore1", "ignore2", "phones"),

    col_types = "cddddc"

  ) |> 

  print(n = Inf)

```



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.



```{r, eval = FALSE}

# 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.



[mfa-docs]: https://montreal-forced-aligner.readthedocs.io/en/latest/index.html "Read The Docs page for the MFA"



[mfa-github]: https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner "MFA GitHub page"



[mfa-phon-prob]: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html#silence-probabilities "MFA Dictionary Format"