Hugging Face's logo

proxectonos
/
Nos_StyleTTS2-Brais-GL

Text-to-Speech
Galician
tts
styletts2
galician
style-tts2
phoneme-level
proxecto-nos
Model card Files
xet
 Community
Nos_StyleTTS2-Brais-GL / Utils /cotovia /docs /tech_documentation.md
cmagui's picture
cmagui
Initial commit: full repository with code, configs and weights
613ce86
preview code
|
Raw
History Blame Contribute Delete
3.41 kB

Technical Documentation

Data

The projects requires additional runtime data to be able to performs its operations. These are the dictionaries in csv format:

  • data/lang/es: Spanish dictionaries.

  • data/lang/gl: Gallego dictionaries.

  • data/lang/gl/abr.txt: Contains a series of abbreviations, along with their expanded form.

  • data/lang/gl/acron.txt: Contains a series of acronyms, along with their expanded form.

  • data/lang/gl/adxectivos.txt: Contains a series of adjectives, along with their number and gender.

  • data/lang/gl/bigramas.dat: Binary data of a 2-gram neural network.

  • data/lang/gl/bigramas_real.dat: Binary data of a 2-gram neural network.

  • data/lang/gl/clases_ambiguedad.dat: Binary data of a neural network that collapses the grammar category of a word.

  • data/lang/gl/derivativos.txt: Contain rules for generating derivative words from certain rules.

  • data/lang/gl/desinencias.txt: Contains a series of verb endings.

  • data/lang/gl/locucions.txt: Contains a series of verb locucions, along with their grammar category.

  • data/lang/gl/nomes.txt: Contains a series nouns, with their number and gender.

  • data/lang/gl/palabrasFunction.txt: Dictionary with all the function words.

  • data/lang/gl/pentagramas.dat: Binary data of a 5-gram neural network.

  • data/lang/gl/pentagramas_real.dat: Binary data of a 5-gram neural network.

  • data/lang/gl/perifrase.txt: Dictionary with all the periphrases.

  • data/lang/gl/tetragramas.dat: Binary data of a 4-gram neural network.

  • data/lang/gl/tetragramas_real.dat: Binary data of a 4-gram neural network.

  • data/lang/gl/trigramas.dat: Binary data of a 3-gram neural network.

  • data/lang/gl/trigramas.dat: Binary data of a 3-gram neural network.

  • data/lang/gl/unigramas.dat: Binary data of a 1-gram neural network.

  • data/lang/gl/utf8chars.csv: Contains a series of custom transformation from UTF-8 words to latin1

  • data/lang/gl/verbos.txt: Contains all the verbs, along with their roots.

Classes

  • Cotovia: The primary entry point of the application. It handles the loading and unloading of dictionaries and processes sentences.
  • PreprocessorUTF8::Processor: Handles the transformation of strings in UTF-8 into latin1.
  • Tokenizar: Manages the tokenization of an input string
  • Preproceso: Applies a series of transformations to a tokenized input string to ensure it is in a canonical form.
  • Analisis_morfologico: Applies a morphologic analysis in which it deduces the classification of each token without contextual information.
  • Analisis_morfosintactico: Uses a neural network and an n-gram model to compute the classification of each word.
  • Sintagma: Performs syntagma analysis on the input tokens, clustering the tokens into groups.
  • Transcripcion: Takes a tokenized sentence and transforms it into multiple phonemes.

Tests

In the "tests" folder, there is a python script that performs integration tests on the cotovia projects. The following aspects are tested:

  • Allomorphs of articles in affirmative sentences with a period
  • Allomorphs of articles in affirmative sentences
  • Allomorphs of articles in exclamatory sentences
  • Allomorphs of articles in interrogative sentences
  • Galician recognition
  • Galician voice
  • Feminine names
  • Masculine names
  • Number recognition
  • Number pronunciation
  • Ordinal numbers
  • Roman numerals
  • Punctuation marks
  • Punctuation marks (variant N)
  • Phonemes