Add a proper phonetic trancription for Japanese

autosyl/LyricsAlignment/utils.py

@@ -5,15 +5,37 @@ import numpy as np
 import librosa
 import string
 import warnings
-from g2p_en import G2p
+import g2p_en
+from g2p import make_g2p
+
 
-g2p = G2p()
 
 phone_dict = ['AA', 'AE', 'AH', 'AO', 'AW', 'AY', 'B', 'CH', 'D', 'DH', 'EH', 'ER', 'EY', 'F', 'G', 'HH', 'IH', 'IY',
              'JH', 'K', 'L', 'M', 'N', 'NG', 'OW', 'OY', 'P', 'R', 'S', 'SH', 'T', 'TH', 'UH', 'UW', 'V', 'W', 'Y',
              'Z', 'ZH', ' ']
 phone2int = {phone_dict[i]: i for i in range(len(phone_dict))}
 
+
+class G2p_Wrapper():
+
+    def __init__(self, language="jp"):
+        if language == "en":
+            self.transducer = g2p_en.G2p()
+        else:                                                   # Only Japanese Romaji for now...
+            self.transducer = make_g2p('rji', 'rji-eng-arpa')
+
+        self.language = language
+
+    def __call__(self, word):
+        if self.language == "en":
+            return self.transducer(word)
+        else:
+            return self.transducer(word).output_string.split()
+
+g2p = G2p_Wrapper(language="jp")
+
+
+
 def my_collate(batch):
     audio, targets, seqs = zip(*batch)
     audio = np.array(audio)

autosyl/segment.py

@@ -37,6 +37,8 @@ def segment(songfile, reference_syls=None, syls_per_line=10, last_syl_dur=500, v
     # idx_word_p:   indices of word start in lyrics_p
     # idx_line_p:   indices of line start in lyrics_p
     audio, words, lyrics_p, idx_word_p, idx_line_p = preprocess_from_file(audio_file, lyrics_lines, word_file)
+    if verbose:
+        print(lyrics_p)
 
     if verbose:
         print("Retrieving syls from lyrics...")

g2p/mappings/langs/rji/config.yaml

+<<: &shared
+  - language_name: Romaji
+mappings:
+  - display_name: Romaji (Hepburn) to English Arpabet
+    in_lang: rji
+    out_lang: rji-eng-arpa
+    authors:
+      - Loïc Allègre
+    type: mapping
+    mapping: romaji_to_eng-arpa.csv
+    abbreviations: rji_abbs.csv
+    rule_ordering: as-written
+    case_sensitive: false
+    norm_form: 'NFC'
+    prevent_feeding: true
+    <<: *shared
\ No newline at end of file

g2p/mappings/langs/rji/rji_abbs.csv

+VOWEL,a,e,i,o,u
+EI_VOW,e,i
+AOU_VOW,a,o,u
\ No newline at end of file

g2p/mappings/langs/rji/romaji_to_eng-arpa.csv

+ch,CH ,,,true
+ou,OW ,,,true
+sh,SH ,,,true
+dj,JH ,,,true
+dz,Z ,,,true
+a,AA ,,,false
+e,EH ,,,false
+i,IY ,,,false
+o,OW ,,,false
+u,UW ,,,false
+k,K ,,y,false
+k,K ,,VOWEL,false
+k,KUW ,,CONS,false
+k,K ,,,false
+s,S ,,VOWEL,false
+s,SUW ,,CONS,false
+s,S ,,,false
+t,T ,,,false
+n,N ,,,false
+h,H ,,,false
+m,M ,,,false
+y,Y ,,,false
+r,L ,,,false
+w,W ,,,false
+g,G ,,,false
+z,Z ,,,false
+d,D ,,,false
+b,B ,,,false
+p,P ,,,false
+f,F ,,,false
+v,V ,,,false
+j,JH ,,,false
+q,K ,,,false
+l,L ,,,false
\ No newline at end of file