From 1ced59bfc0b86295ab837862352ca420fd80d61e Mon Sep 17 00:00:00 2001
From: Brandon Thomas <echelon@gmail.com>
Date: Tue, 5 Jul 2022 01:59:21 -0400
Subject: [PATCH] require instance of phonetization to be passed

---
 tts/text/pipeline/english_pipeline_v1.py      |  9 +++++----
 tts/text/pipeline/test_english_pipeline_v1.py | 20 +++++++++++++------
 2 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/tts/text/pipeline/english_pipeline_v1.py b/tts/text/pipeline/english_pipeline_v1.py
index 790087b..3af6ed9 100644
--- a/tts/text/pipeline/english_pipeline_v1.py
+++ b/tts/text/pipeline/english_pipeline_v1.py
@@ -2,13 +2,11 @@ from typing import List
 from tts.text.pipeline.base_pipeline import BasePipeline
 from tts.text.pipeline_components.input_parsing.parser import ArpabetSequence, GraphemeSequence, ParsedSequence, parse_grapheme_and_arpabet_sequence
 from tts.text.pipeline_components.normalization.english_normalization import normalize_english
+from tts.text.pipeline_components.phonetization.base_phonetization import BasePhonetization
 from tts.text.pipeline_components.phonetization.cmudict import DEFAULT_CMUDICT_PATH, CMUDict
 from tts.text.pipeline_components.phonetization.english_phonetization import EnglishPhonetization
 from tts.text.pipeline_components.sequencing.english_symbols_v1 import SPACE_INTEGER_ENCODING, arpabet_to_sequence, symbols_to_sequence
 
-_ENGLISH_DICTIONARY = CMUDict(DEFAULT_CMUDICT_PATH)
-_ENGLISH_PHONETIZER = EnglishPhonetization(_ENGLISH_DICTIONARY)
-
 class EnglishPipelineV1(BasePipeline):
     """
     This is a new text pipeline for FakeYou that will be introduced in July 2022.
@@ -17,6 +15,9 @@ class EnglishPipelineV1(BasePipeline):
     the model input expectations on both sides match one another.
     """
 
+    def __init__(self, phonetization: BasePhonetization):
+        self.phonetization = phonetization
+
     def user_input_to_sequence(self, input_text: str) -> List[int]:
         """Convert raw user input into a numerical sequence for ML inference."""
 
@@ -45,7 +46,7 @@ class EnglishPipelineV1(BasePipeline):
                 updated_sequence.append(subsequence)
 
             elif isinstance(subsequence, GraphemeSequence):
-                converted_sequence = _ENGLISH_PHONETIZER.try_convert_to_phonemes(subsequence.sequence)
+                converted_sequence = self.phonetization.try_convert_to_phonemes(subsequence.sequence)
                 updated_sequence.extend(converted_sequence)
 
         return updated_sequence
diff --git a/tts/text/pipeline/test_english_pipeline_v1.py b/tts/text/pipeline/test_english_pipeline_v1.py
index 0c81128..6bec91c 100644
--- a/tts/text/pipeline/test_english_pipeline_v1.py
+++ b/tts/text/pipeline/test_english_pipeline_v1.py
@@ -1,25 +1,33 @@
 from tts.text.pipeline.english_pipeline_v1 import EnglishPipelineV1
+from tts.text.pipeline_components.phonetization.cmudict import DEFAULT_CMUDICT_PATH, CMUDict
+from tts.text.pipeline_components.phonetization.english_phonetization import EnglishPhonetization
 
 
 class TestEnglishPipelineV1:
 
+    def build_pipeline_instance(self):
+        # NB: CMUDict used in production may differ
+        cmudict = CMUDict(DEFAULT_CMUDICT_PATH)
+        phonetization = EnglishPhonetization(cmudict)
+        return EnglishPipelineV1(phonetization)
+
     def test_no_input(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("")
         assert output == []
 
     def test_whitespace(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("   \n    \t ")
         assert output == []
 
     def test_single_word(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("hello")
         assert output == [106, 73, 117, 123] # NB: Arpabet encoded
 
     def test_two_words(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("hello hello")
         assert output == [
             106, 73, 117, 123, # NB: Arpabet encoded
@@ -28,7 +36,7 @@ class TestEnglishPipelineV1:
         ]
 
     def test_two_words_with_arpabet_segment(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("hello {HH AH0 L OW1}")
         assert output == [
             106, 73, 117, 123, # NB: Arpabet encoded
@@ -37,7 +45,7 @@ class TestEnglishPipelineV1:
         ]
 
     def test_words_and_punctuation(self):
-        pipeline = EnglishPipelineV1()
+        pipeline = self.build_pipeline_instance()
         output = pipeline.user_input_to_sequence("hello friends, welcome!")
         assert output == [
             106, 73, 117, 123, # NB: Arpabet encoded