Merge pull request #515 from EveryVoiceTTS/dev.ap/text-fixes

fix: change shape of filelist list data instead of re-reading it

everyvoice/config/text_config.py

@@ -5,7 +5,7 @@
 from everyvoice.config.shared_types import ConfigModel
 from everyvoice.config.utils import PossiblySerializedCallable
 from everyvoice.text.utils import normalize_text_helper
-from everyvoice.utils import collapse_whitespace
+from everyvoice.utils import collapse_whitespace, strip_text
 
 
 class Punctuation(BaseModel):
@@ -62,6 +62,23 @@ def all_except_punctuation(self) -> set[str]:
         """Returns the set containing all characters."""
         return set(w for _, v in self if not isinstance(v, Punctuation) for w in v)
 
+    @model_validator(mode="after")
+    def cannot_have_punctuation_in_symbol_set(self) -> "Symbols":
+        """You cannot have the same symbol defined in punctuation as elsewhere.
+
+        Raises:
+            ValueError: raised if a symbol from punctuation is found elsewhere
+
+        Returns:
+            Symbols: The validated symbol set
+        """
+        for punctuation in self.punctuation.all:
+            if punctuation in self.all_except_punctuation:
+                raise ValueError(
+                    f"Sorry, the symbol '{punctuation}' occurs in both your declared punctuation and in your other symbol set. Please inspect your text configuration and either remove the symbol from the punctuation or other symbol set."
+                )
+        return self
+
     @model_validator(mode="after")
     def member_must_be_list_of_strings(self) -> "Symbols":
         """Except for `punctuation` & `pad`, all user defined member variables
@@ -81,9 +98,7 @@ def member_must_be_list_of_strings(self) -> "Symbols":
 class TextConfig(ConfigModel):
     symbols: Symbols = Field(default_factory=Symbols)
     to_replace: Dict[str, str] = {}  # Happens before cleaners
-    cleaners: list[PossiblySerializedCallable] = [
-        collapse_whitespace,
-    ]
+    cleaners: list[PossiblySerializedCallable] = [collapse_whitespace, strip_text]
 
     @model_validator(mode="after")
     def clean_symbols(self) -> "TextConfig":

everyvoice/model/e2e/config/__init__.py

@@ -18,15 +18,15 @@
 
 # The contact information only needs to be registered on the main config
 class AlignerConfigNoContact(AlignerConfig):
-    contact: Optional[ContactInformation] = None
+    contact: Optional[ContactInformation] = None  # type: ignore
 
 
 class VocoderConfigNoContact(VocoderConfig):
-    contact: Optional[ContactInformation] = None
+    contact: Optional[ContactInformation] = None  # type: ignore
 
 
 class FeaturePredictionConfigNoContact(FeaturePredictionConfig):
-    contact: Optional[ContactInformation] = None
+    contact: Optional[ContactInformation] = None  # type: ignore
 
 
 class E2ETrainingConfig(BaseTrainingConfig):
@@ -36,17 +36,17 @@ class E2ETrainingConfig(BaseTrainingConfig):
 
 class EveryVoiceConfig(BaseModelWithContact):
     aligner: AlignerConfig | AlignerConfigNoContact = Field(
-        default_factory=AlignerConfigNoContact
+        default_factory=AlignerConfigNoContact  # type: ignore
     )
     path_to_aligner_config_file: Optional[FilePath] = None
 
     feature_prediction: FeaturePredictionConfig | FeaturePredictionConfigNoContact = (
-        Field(default_factory=FeaturePredictionConfigNoContact)
+        Field(default_factory=FeaturePredictionConfigNoContact)  # type: ignore
     )
     path_to_feature_prediction_config_file: Optional[FilePath] = None
 
     vocoder: VocoderConfig | VocoderConfigNoContact = Field(
-        default_factory=VocoderConfigNoContact
+        default_factory=VocoderConfigNoContact  # type: ignore
     )
     path_to_vocoder_config_file: Optional[FilePath] = None
 

everyvoice/tests/data/relative/config/everyvoice-shared-text.yaml

@@ -1,6 +1,6 @@
 cleaners: [everyvoice.utils.lower, everyvoice.utils.collapse_whitespace, everyvoice.utils.nfc_normalize]
 symbols:
-  dataset_0-symbols: [' ', '''', ',', '-', ., C, E, H, K, P, T, a, b, c, d, e, f,
+  dataset_0-symbols: [' ', C, E, H, K, P, T, a, b, c, d, e, f,
     g, h, i, l, m, n, o, p, r, s, t, u, v, w, x, y]
   pad: _
   silence: [<SIL>]

everyvoice/tests/data/unit-test-case1.psv

@@ -1,4 +1,5 @@
 columnA|non-standard-basename|non-standard-text|extra
 blah|somefile|characters|irrelevant extra
 boom|file2|CaSeD 	 NFD: éàê NFC: éàê|blah
+floop|file3|  let us  see    if   it collapses whitespace|blah
 bam|banned_file|ZZZ|has banned symbol (Z)

everyvoice/tests/test_text.py

@@ -48,8 +48,8 @@ def test_token_sequence_to_text(self):
 
     def test_hardcoded_symbols(self):
         self.assertEqual(
-            self.base_text_processor.encode_text("\x80 "),
-            [0, 1],
+            self.base_text_processor.encode_text("\x80 \x80"),
+            [0, 1, 0],
             "pad should be Unicode PAD symbol and index 0, whitespace should be index 1",
         )
 
@@ -65,6 +65,10 @@ def test_cleaners_with_upper(self):
         sequence = upper_text_processor.encode_text(text_upper)
         self.assertEqual(upper_text_processor.decode_tokens(sequence, ""), text)
 
+    def test_no_duplicate_punctuation(self):
+        with self.assertRaises(ValidationError):
+            TextConfig(symbols=Symbols(letters=[":"] + list(string.ascii_letters)))
+
     def test_punctuation(self):
         text = "hello! How are you? My name's: foo;."
         upper_text_processor = TextProcessor(

everyvoice/tests/test_wizard.py

@@ -408,6 +408,80 @@ def test_sample_rate_config(self):
         self.assertTrue(step.completed)
         self.assertEqual(step.response, 512)
 
+    def test_whitespace_always_collapsed(self):
+        tour = Tour("unit testing", steps=dataset.get_dataset_steps())
+        filelist = str(self.data_dir / "unit-test-case1.psv")
+        filelist_step = find_step(SN.filelist_step, tour.steps)
+        monkey = monkeypatch(filelist_step, "prompt", Say(filelist))
+        with monkey:
+            filelist_step.run()
+
+        permission_step = find_step(SN.dataset_permission_step, tour.steps)
+        with patch_menu_prompt(1):  # 1 is "yes, I have permission"
+            permission_step.run()
+        self.assertTrue(
+            permission_step.state[SN.dataset_permission_step].startswith("Yes")
+        )
+
+        format_step = find_step(SN.filelist_format_step, tour.steps)
+        with patch_menu_prompt(0):  # 0 is "psv"
+            format_step.run()
+        step = format_step.children[0]
+        with patch_menu_prompt(1):  # 1 is "yes"
+            step.run()
+
+        step = format_step.children[1]
+        with patch_menu_prompt(1):  # 1 is second column
+            step.run()
+
+        step = format_step.children[2]
+        with patch_menu_prompt(1):  # 1 is second remaining column, i.e., third column
+            step.run()
+
+        text_representation_step = find_step(
+            SN.filelist_text_representation_step, tour.steps
+        )
+        with patch_menu_prompt(0):  # 0 is "characters"
+            text_representation_step.run()
+        speaker_step = find_step(SN.data_has_speaker_value_step, tour.steps)
+        with patch_menu_prompt(0):  # 0 is "no"
+            speaker_step.run()
+
+        know_speaker_step = speaker_step.children[0]
+        with patch_menu_prompt(1):  # 1 is "yes"
+            know_speaker_step.run()
+
+        add_speaker_step = know_speaker_step.children[0]
+        with patch_input("default"):
+            add_speaker_step.run()
+
+        language_step = find_step(SN.data_has_language_value_step, tour.steps)
+        with patch_menu_prompt(0):  # 0 is "no"
+            language_step.run()
+
+        select_lang_step = language_step.children[0]
+        with capture_stdout(), capture_stderr():
+            with patch_menu_prompt(15):  # some arbitrary language from the list
+                select_lang_step.run()
+
+        wavs_dir_step = find_step(SN.wavs_dir_step, tour.steps)
+        with monkeypatch(wavs_dir_step, "prompt", Say(str(self.data_dir))):
+            wavs_dir_step.run()
+
+        validate_wavs_step = find_step(SN.validate_wavs_step, tour.steps)
+        with patch_menu_prompt(1), capture_stdout():
+            validate_wavs_step.run()
+
+        text_processing_step = find_step(SN.text_processing_step, tour.steps)
+        # 0 is lowercase, 1 is NFC Normalization, select none
+        with monkeypatch(dataset, "tqdm", lambda seq, desc: seq):
+            with patch_menu_prompt([]):
+                text_processing_step.run()
+        self.assertEqual(
+            text_processing_step.state["filelist_data_list"][3][2],
+            "let us see if it collapses whitespace",
+        )
+
     def test_dataset_subtour(self):
         tour = Tour("unit testing", steps=dataset.get_dataset_steps())
 
@@ -435,7 +509,7 @@ def test_dataset_subtour(self):
         with patch_menu_prompt(1):  # 1 is "yes"
             step.run()
         self.assertEqual(step.state[SN.data_has_header_line_step.value], "yes")
-        self.assertEqual(len(step.state["filelist_data_list"]), 4)
+        self.assertEqual(len(step.state["filelist_data_list"]), 5)
 
         step = format_step.children[1]
         self.assertIsInstance(step, dataset.HeaderStep)
@@ -503,7 +577,7 @@ def test_dataset_subtour(self):
         with patch_menu_prompt(1), capture_stdout() as out:
             validate_wavs_step.run()
         self.assertEqual(step.state[SN.validate_wavs_step][:2], "No")
-        self.assertIn("Warning: 3 wav files were not found", out.getvalue())
+        self.assertIn("Warning: 4 wav files were not found", out.getvalue())
 
         text_processing_step = find_step(SN.text_processing_step, tour.steps)
         # 0 is lowercase, 1 is NFC Normalization, select both
@@ -513,11 +587,16 @@ def test_dataset_subtour(self):
         # print(text_processing_step.state)
         self.assertEqual(
             text_processing_step.state["filelist_data_list"][2][2],
-            "cased \t nfd: éàê nfc: éàê",  # the "nfd: éàê" bit here is now NFC
+            "cased nfd: éàê nfc: éàê",  # the "nfd: éàê" bit here is now NFC
+        )
+
+        self.assertEqual(
+            text_processing_step.state["filelist_data_list"][3][2],
+            "let us see if it collapses whitespace",
         )
 
         # Make sure realoading the data as dict stripped the header line
-        self.assertEqual(len(step.state["filelist_data"]), 3)
+        self.assertEqual(len(step.state["filelist_data"]), 4)
 
         sox_effects_step = find_step(SN.sox_effects_step, tour.steps)
         # 0 is resample to 22050 kHz, 2 is remove silence at start and end
@@ -547,11 +626,24 @@ def test_dataset_subtour(self):
         )
 
         symbol_set_step = find_step(SN.symbol_set_step, tour.steps)
-        self.assertEqual(len(symbol_set_step.state["filelist_data"]), 3)
+        self.assertEqual(len(symbol_set_step.state["filelist_data"]), 4)
         with capture_stdout(), capture_stderr():
             symbol_set_step.run()
         self.assertEqual(len(symbol_set_step.state[SN.symbol_set_step.value]), 2)
         self.assertIn("t͡s", symbol_set_step.state[SN.symbol_set_step.value]["phones"])
+        self.assertNotIn(
+            ":", symbol_set_step.state[SN.symbol_set_step.value]["characters"]
+        )
+        self.assertNotIn(":", symbol_set_step.state[SN.symbol_set_step.value]["phones"])
+        # assert that symbols contain no duplicates
+        self.assertEqual(
+            len(set(symbol_set_step.state[SN.symbol_set_step.value]["characters"])),
+            len(symbol_set_step.state[SN.symbol_set_step.value]["characters"]),
+        )
+        self.assertEqual(
+            len(set(symbol_set_step.state[SN.symbol_set_step.value]["phones"])),
+            len(symbol_set_step.state[SN.symbol_set_step.value]["phones"]),
+        )
 
     def test_empty_filelist(self):
         tour = Tour(
@@ -1599,8 +1691,6 @@ def setUp(self):
                         SN.symbol_set_step.value: {
                             "characters": [
                                 " ",
-                                ",",
-                                ".",
                                 "A",
                                 "D",
                                 "E",

everyvoice/utils/__init__.py

@@ -356,14 +356,22 @@ def generic_dict_loader(
 generic_csv_filelist_reader = partial(generic_dict_loader, delimiter=",")
 
 
-def collapse_whitespace(text):
+def collapse_whitespace(text: str):
     """
     >>> collapse_whitespace("  asdf  	   qwer   ")
     ' asdf qwer '
     """
     return re.sub(_whitespace_re, " ", text)
 
 
+def strip_text(text: str):
+    """
+    >>> strip_text("  asdf  qwer ")
+    'asdf  qwer'
+    """
+    return text.strip()
+
+
 @contextmanager
 def tqdm_joblib_context(tqdm_instance):
     """Context manager to make tqdm compatible with joblib.Parallel