Aligner

• Montreal Forced Aligner
• Montreal Forced Aligner - Github
• Acoustic models
• Pronunciation dictionaries

G2P Model

Output information about the G2P model.

mfa \
  model \
  inspect \
  acoustic \
  french_qc

- Acoustic model: french_qc
  - Version: 1.0.0
  - Architecture: gmm-hmm
  - Phone type: triphone
  - Features:
    - Feature type: mfcc
    - Frame shift: 10
    - Performs speaker adaptation: False
    - Performs LDA on features: False

  - Phones: @, E, N, O, R, S, Z, ^,
            a, b, cinq, d, deux, e, f, g,
            huit, i, j, k, l, m, n, neuf,
            o, p, s, t, to, u, un, v,
            w, y, and z
  - Configuration options:
    - Multilingual IPA: False

Validate

mfa \
  validate \
    ~/mfa_data/my_corpus \
    french_ipa \
    french_qc \
    ~/mfa_data/my_corpus_aligned

Align

mfa \
  align \
    ~/mfa_data/my_corpus \
    french_ipa \
    french_qc \
    ~/mfa_data/my_corpus_aligned

docker container run --rm -it --volume $PWD/Librispeech:/Librispeech common-voice_mfa mfa validate /Librispeech/19  /Librispeech/librispeech-lexicon.txt french_qc /Librispeech/test