From 6de4ee61a03d3f5f8184195c92e58b7f5b8ba642 Mon Sep 17 00:00:00 2001
From: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Date: Tue, 19 Apr 2022 13:39:04 +0200
Subject: [PATCH] Wav2 vec2 phoneme ctc tokenizer optimisation (#16817)

* Solved href rendering issue in heading

Markdown references in headings such as '####' don't render well.
Replaced it with <h4>...<a></a></h> banners.

* PhonemeTokenizer optimization using phonemizer lib

The backend should only be initialized once, otherwise it is reloaded.
Added `init_backend` function, intializes a backend attribute.
Phonemize re-uses self.backend.
Should give ~10 times faster phonemization.

* formatted file with make style

* Documentation suggestion

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update /tokenization_wav2vec2_phoneme.py based on PR suggestion

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update CONTRIBUTING.md

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
---
 CONTRIBUTING.md                               |  3 +-
 .../tokenization_wav2vec2_phoneme.py          | 33 +++++++++++++------
 2 files changed, 24 insertions(+), 12 deletions(-)

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index c0af3fbaa6..e74510948a 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -368,8 +368,7 @@ For documentation strings, 🤗 Transformers follows the [google style](https://
 Check our [documentation writing guide](https://github.com/huggingface/transformers/tree/main/docs#writing-documentation---specification)
 for more information.
 
-#### This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md)
-
+**This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md).**
 
 ### Develop on Windows
 
diff --git a/src/transformers/models/wav2vec2_phoneme/tokenization_wav2vec2_phoneme.py b/src/transformers/models/wav2vec2_phoneme/tokenization_wav2vec2_phoneme.py
index b37f902c34..6bd355645e 100644
--- a/src/transformers/models/wav2vec2_phoneme/tokenization_wav2vec2_phoneme.py
+++ b/src/transformers/models/wav2vec2_phoneme/tokenization_wav2vec2_phoneme.py
@@ -158,6 +158,9 @@ class Wav2Vec2PhonemeCTCTokenizer(PreTrainedTokenizer):
         self.phonemizer_lang = phonemizer_lang
         self.phonemizer_backend = phonemizer_backend
 
+        if do_phonemize:
+            self.init_backend(self.phonemizer_lang)
+
         with open(vocab_file, encoding="utf-8") as vocab_handle:
             self.encoder = json.load(vocab_handle)
         self.decoder = {v: k for k, v in self.encoder.items()}
@@ -169,6 +172,18 @@ class Wav2Vec2PhonemeCTCTokenizer(PreTrainedTokenizer):
     def get_vocab(self) -> Dict:
         return dict(self.encoder, **self.added_tokens_encoder)
 
+    def init_backend(self, phonemizer_lang: str):
+        """
+        Initializes the backend.
+
+        Args:
+            phonemizer_lang (`str`): The language to be used.
+        """
+        requires_backends(self, "phonemizer")
+        from phonemizer.backend import BACKENDS
+
+        self.backend = BACKENDS[self.phonemizer_backend](phonemizer_lang, language_switch="remove-flags")
+
     def prepare_for_tokenization(
         self,
         text: str,
@@ -209,6 +224,7 @@ class Wav2Vec2PhonemeCTCTokenizer(PreTrainedTokenizer):
         # set the correct phonemizer language
         if phonemizer_lang is not None:
             self.phonemizer_lang = phonemizer_lang
+            self.init_backend(phonemizer_lang)
 
         return (text, {})
 
@@ -234,23 +250,20 @@ class Wav2Vec2PhonemeCTCTokenizer(PreTrainedTokenizer):
         return tokens
 
     def phonemize(self, text: str, phonemizer_lang: Optional[str] = None) -> str:
-        requires_backends(self, "phonemizer")
-
-        from phonemizer import phonemize
         from phonemizer.separator import Separator
 
         word_delimiter = self.word_delimiter_token + " " if self.word_delimiter_token is not None else ""
-        phonemizer_lang = phonemizer_lang if phonemizer_lang is not None else self.phonemizer_lang
+        if phonemizer_lang is not None and phonemizer_lang != self.phonemizer_lang:
+            self.init_backend(phonemizer_lang)
+        else:
+            phonemizer_lang = self.phonemizer_lang
 
         separator = Separator(phone=self.phone_delimiter_token, word=word_delimiter, syllable="")
-        phonemes = phonemize(
-            text,
-            language=phonemizer_lang,
-            backend=self.phonemizer_backend,
+        phonemes = self.backend.phonemize(
+            [text],
             separator=separator,
-            language_switch="remove-flags",
         )
-        phonemes = phonemes.strip()
+        phonemes = phonemes[0].strip()
 
         return phonemes