Espeak phonemizer pipeline for spanish (and also english)

2026-10-25 08:40:31 +00:00 · 2022-07-14 22:14:32 -03:00
parent 2e6be74403
commit 840a97e70e
13 changed files with 491 additions and 15 deletions
@@ -0,0 +1,127 @@
+import re
+
+from phonemizer.backend import EspeakBackend
+from phonemizer.separator import Separator
+
+from typing import List
+from text.pipeline_components.input_parsing.parser import ArpabetSequence, GraphemeSequence, ParsedSequence
+from text.pipeline_components.phonetization.base_phonetization import BasePhonetization
+from text.pipeline_components.input_parsing.parser import parse_grapheme_and_arpabet_sequence
+# IPA to ARPABET
+
+basic_replacements = {
+  'aɪ': 'AA0 Y',
+  'aʊ': 'AA0 W',
+  'b': 'B',
+  'd': 'D',
+  'dʒ': 'JH',
+  'eɪ': 'EY0',
+  'f': 'F',
+  'h': 'HH',
+  'i': 'IY0',
+  'j': 'Y',
+  'k': 'K',
+  'l': 'L',
+  'l̩': 'EL',
+  'm': 'M',
+  'm̩': 'EM',
+  'n': 'N',
+  'n̩': 'EN',
+  'oʊ': 'OW',
+  'p': 'P',
+  's': 'S',
+  't': 'T',
+  'tʃ': 'CH',
+  'u': 'UW0',
+  'v': 'V',
+  'w': 'W',
+  'z': 'Z',
+  'æ': 'AE0',
+  'ð': 'DH',
+  'ŋ': 'NG',
+  'ɑ': 'AA0',
+  'ɔ': 'AO0',
+  'ɔɪ': 'OY0',
+  'ə': 'AX',
+  'ɚ': 'AXR',
+  'ɛ': 'EH0',
+  'ɝ': 'ER',
+  'ɡ': 'G',
+  'ɨ': 'IX',
+  'ɪ': 'IH0',
+  'ɹ': 'R',
+  'ɾ': 'R',
+  'ɾ̃': 'NX',
+  'ʃ': 'SH',
+  'ʉ': 'UX',
+  'ʊ': 'UH0',
+  'ʌ': 'AH0',
+  'ʍ': 'WH',
+  'ʒ': 'ZH',
+  'ʔ': 'Q',
+  'θ': 'TH'
+}
+
+class EspeakPhonetization(BasePhonetization):
+
+    def __init__(self, lang="en"):
+        super().__init__(None)
+        if lang == 'en':
+            from text.pipeline_components.phonetization.lang.es import valid_symbols, extra_replacements
+        elif lang == 'es':
+            from text.pipeline_components.phonetization.lang.es import valid_symbols, extra_replacements
+        else:
+            raise Exception("Not supported language")
+
+
+        self.replacements = {**basic_replacements, **extra_replacements}
+        self.separator = Separator(phone='|', word=None)
+
+        lang2espeaklang = {"en": "en-us", "es": "es", "fr": "fr-fr"}
+        self.backend = EspeakBackend(lang2espeaklang[lang],with_stress=True, preserve_punctuation=False)
+
+    def grapheme2ipa(self, word):
+        return self.backend.phonemize([word],separator=self.separator)[0].strip().strip("|")
+
+    def ipa2arpabet(self, ipa_word):
+        # print(ipa_word)
+        ipa_word = ipa_word.replace("ɾ|ɾ", "r")
+        if ipa_word.endswith("j|j"):
+            ipa_word = ipa_word.replace("j|j", "i")
+        else:
+            ipa_word = ipa_word.replace("j|j", "ʎ")
+        # Remove secondary accents, they are not important in spanish
+        ipa_word = ipa_word.replace("ˌ", "")
+        # Remove long consonant indication
+        ipa_word = ipa_word.replace("ː", "")
+        arpabet_word = ""
+        for c in ipa_word.split("|"):
+            accent = False
+            if c.startswith("ˈ"):
+                c = c[1:]
+                accent = True
+            if c in self.replacements:
+                replacement = self.replacements[c]
+                if accent:
+                    replacement = replacement.replace("0", "1")
+                arpabet_word += replacement + " "
+            else:
+                print(ipa_word)
+                print(f"'{c}' is not in the replacements dictionary")
+                raise Exception()
+        return "{" + arpabet_word.strip() + "}"
+
+    def word2phonemes(self, word):
+        return self.ipa2arpabet(self.grapheme2ipa(word))
+
+    def __len__(self):
+        return len(self._entries)
+
+    def try_convert_to_phonemes(self, grapheme_sequence: str) -> ParsedSequence:
+        '''Converts word to spanish adapted ARPAbet'''
+
+        # Apply it only on words, keep punctuation unchanged
+        for match in re.findall(r"[A-Za-zÀ-ÿ]+", grapheme_sequence):
+            grapheme_sequence = grapheme_sequence.replace(match, self.word2phonemes(match), 1)
+
+        return parse_grapheme_and_arpabet_sequence(grapheme_sequence)
@@ -0,0 +1,19 @@
+
+_pad        = '_'
+_punctuation = '!\'(),.:;? ' #NB: Includes space
+_special = '-'
+_letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
+
+# https://en.wikipedia.org/wiki/ARPABET
+valid_symbols = [
+  'AA', 'AA0', 'AA1', 'AA2', 'AE', 'AE0', 'AE1', 'AE2', 'AH', 'AH0', 'AH1', 'AH2',
+  'AO', 'AO0', 'AO1', 'AO2', 'AW', 'AW0', 'AW1', 'AW2', 'AY', 'AY0', 'AY1', 'AY2',
+  'B', 'CH', 'D', 'DH', 'EH', 'EH0', 'EH1', 'EH2', 'ER', 'ER0', 'ER1', 'ER2', 'EY',
+  'EY0', 'EY1', 'EY2', 'F', 'G', 'HH', 'IH', 'IH0', 'IH1', 'IH2', 'IY', 'IY0', 'IY1',
+  'IY2', 'JH', 'K', 'L', 'M', 'N', 'NG', 'OW', 'OW0', 'OW1', 'OW2', 'OY', 'OY0',
+  'OY1', 'OY2', 'P', 'R', 'S', 'SH', 'T', 'TH', 'UH', 'UH0', 'UH1', 'UH2', 'UW',
+  'UW0', 'UW1', 'UW2', 'V', 'W', 'Y', 'Z', 'ZH'
+]
+
+extra_replacements = {}
+
@@ -0,0 +1,34 @@
+# https://en.wikipedia.org/wiki/ARPABET
+valid_symbols = [
+  'AA', 'AA0', 'AA1', 'AA2', 'AE', 'AE0', 'AE1', 'AE2', 'AH', 'AH0', 'AH1', 'AH2',
+  'AO', 'AO0', 'AO1', 'AO2', 'AW', 'AW0', 'AW1', 'AW2', 'AY', 'AY0', 'AY1', 'AY2',
+  'B', 'CH', 'D', 'DH', 'EH', 'EH0', 'EH1', 'EH2', 'ER', 'ER0', 'ER1', 'ER2', 'EY',
+  'EY0', 'EY1', 'EY2', 'F', 'G', 'HH', 'IH', 'IH0', 'IH1', 'IH2', 'IY', 'IY0', 'IY1',
+  'IY2', 'JH', 'K', 'L', 'M', 'N', 'NG', 'OW', 'OW0', 'OW1', 'OW2', 'OY', 'OY0',
+  'OY1', 'OY2', 'P', 'R', 'S', 'SH', 'T', 'TH', 'UH', 'UH0', 'UH1', 'UH2', 'UW',
+  'UW0', 'UW1', 'UW2', 'V', 'W', 'Y', "RR", "NI"
+]
+
+extra_replacements = {
+    "a": "AA0",
+    "o": "OW0",
+    "β": "V",
+    "e": "EY0",
+    "ɲ": "NI",
+    "r": "RR",
+    "ɣ": "G",
+    "ʎ": "SH",
+    "x": "HH",
+    "ts": "TH", # jacuzzi / pizza
+    "oɪ": "OW0 Y", # acoitar
+    "eʊ": "EY0 W", # adeuda
+    "ʝ": "SH" # ya
+}
+
+# NB: DO NOT CHANGE THE ORDER OF SYMBOLS! 
+# Models are trained on and depend upon the order.
+pad        = '_'
+punctuation = '!\'(),.:;? '
+special = '-'
+# Little hack, since I'm using arpabet letters are not useful anymore, I just replace it with spanish symbols that I need like ¡ and ¿
+letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwx¡¿'