search

v-nhandt21 / ViMFA

Montreal Forced Aligner for Vietnamese
MIT License
7 stars 2 forks source link

Vowels and consonants in viIPA. #2

Open drlor2k opened 3 months ago

drlor2k commented 3 months ago

hello @v-nhandt21 , I have a question, please help me!

Based on the following quote:

A universal monosyllabic phoneme system has "C(m)-V-C(n)" (m,n >= 0) phoneme patterns.

Based on the MFA author's dictionary I was able to find a list of vowels and consonants.

With viIPA, do you have a specific list for vowels and consonants?

v-nhandt21 commented 3 months ago

hello @v-nhandt21 , I have a question, please help me!

Based on the following quote:

A universal monosyllabic phoneme system has "C(m)-V-C(n)" (m,n >= 0) phoneme patterns.

Based on the MFA author's dictionary I was able to find a list of vowels and consonants.

With viIPA, do you have a specific list for vowels and consonants?

Yep, for Viphoneme, you can get list of phoneme by: https://pypi.org/project/viphoneme/1.0.5/

image

drlor2k commented 3 months ago

hi @v-nhandt21, I have some questions:

  1. Each phonetic word in your IPA system has only one vowel, right?
  2. If I merge the tone with the vowel to create another viIPA.txt (ie bận -> b ɤ̆ n 6 -> b ɤ̆6 n), will the whole process of viMFA be affected?
  3. Because the author's MFA for Vietnamese is quite limited in terms of vocabulary, I want to train on my data set. I see version 3.0 of the original author training on a dataset of about 40 hours. However, my dataset is only about 6 hours. Does large and small data affect the final TextGrid quality?

Thank you :3

v-nhandt21 commented 3 months ago

hi @v-nhandt21, I have some questions:

  1. Each phonetic word in your IPA system has only one vowel, right?
  2. If I merge the tone with the vowel to create another viIPA.txt (ie bận -> b ɤ̆ n 6 -> b ɤ̆6 n), will the whole process of viMFA be affected?
  3. Because the author's MFA for Vietnamese is quite limited in terms of vocabulary, I want to train on my data set. I see version 3.0 of the original author training on a dataset of about 40 hours. However, my dataset is only about 6 hours. Does large and small data affect the final TextGrid quality?

Thank you :3

Hi @drlor2k ,

  1. Yep :))
  2. I think "b ɤ̆6 n" seems to be more effective than "b ɤ̆ n 6"
  3. I am not sure about the size and quality of your dataset, hihi, so try to use as much as possible.