Merge pull request #256 from fireblade2534/Fixing-number-normalization

README.md

@@ -516,7 +516,36 @@ Monitor system state and resource usage with these endpoints:
 Useful for debugging resource exhaustion or performance issues.
 </details>
 
-## Known Issues
+## Known Issues & Troubleshooting
+
+<details>
+<summary>Missing words & Missing some timestamps</summary>
+
+The api will automaticly do text normalization on input text which may incorrectly remove or change some phrases. This can be disabled by adding `"normalization_options":{"normalize": false}` to your request json:
+```python
+import requests
+
+response = requests.post(
+    "http://localhost:8880/v1/audio/speech",
+    json={
+        "input": "Hello world!",
+        "voice": "af_heart",
+        "response_format": "pcm",
+        "normalization_options":
+        {
+            "normalize": False
+        }
+    },
+    stream=True
+)
+
+for chunk in response.iter_content(chunk_size=1024):
+    if chunk:
+        # Process streaming chunks
+        pass
+```
+  
+</details>
 
 <details>
 <summary>Versioning & Development</summary>

api/src/routers/openai_compatible.py

@@ -125,20 +125,18 @@ async def process_and_validate_voices(voice_input: Union[str, List[str]], tts_se
 async def stream_audio_chunks(tts_service: TTSService, request: Union[OpenAISpeechRequest, CaptionedSpeechRequest], client_request: Request, writer: StreamingAudioWriter) -> AsyncGenerator[AudioChunk, None]:
     """Stream audio chunks as they're generated with client disconnect handling"""
     voice_name = await process_and_validate_voices(request.voice, tts_service)
-
     unique_properties = {"return_timestamps": False}
     if hasattr(request, "return_timestamps"):
         unique_properties["return_timestamps"] = request.return_timestamps
 
     try:
-        logger.info(f"Starting audio generation with lang_code: {request.lang_code}")
         async for chunk_data in tts_service.generate_audio_stream(
             text=request.input,
             voice=voice_name,
             writer=writer,
             speed=request.speed,
             output_format=request.response_format,
-            lang_code=request.lang_code or settings.default_voice_code or voice_name[0].lower(),
+            lang_code=request.lang_code,
             normalization_options=request.normalization_options,
             return_timestamps=unique_properties["return_timestamps"],
         ):

api/src/services/streaming_audio_writer.py

@@ -25,7 +25,7 @@ class StreamingAudioWriter:
         if self.format in ["wav","flac","mp3","pcm","aac","opus"]:
             if self.format != "pcm":
                 self.output_buffer = BytesIO()
-                self.container = av.open(self.output_buffer, mode="w", format=self.format)
+                self.container = av.open(self.output_buffer, mode="w", format=self.format if self.format != "aac" else "adts")
                 self.stream = self.container.add_stream(codec_map[self.format],sample_rate=self.sample_rate,layout='mono' if self.channels == 1 else 'stereo')
                 self.stream.bit_rate = 128000
         else:

api/src/services/text_processing/normalizer.py

@@ -8,9 +8,11 @@ import re
 from functools import lru_cache
 import inflect
 from numpy import number
-
+from torch import mul
 from ...structures.schemas import NormalizationOptions
 
+from text_to_num import text2num
+
 # Constants
 VALID_TLDS = [
     "com",
@@ -134,25 +136,35 @@ def handle_units(u: re.Match[str]) -> str:
         unit[0]=INFLECT_ENGINE.no(unit[0],number)
     return " ".join(unit)
 
+def conditional_int(number: float, threshold: float = 0.00001):
+    if abs(round(number) - number) < threshold:
+        return int(round(number))
+    return number
+
 def handle_money(m: re.Match[str]) -> str:
     """Convert money expressions to spoken form"""
-    m = m.group()
-    bill = "dollar" if m[0] == "$" else "pound"
-    if m[-1].isalpha():
-        return f"{INFLECT_ENGINE.number_to_words(m[1:])} {bill}s"
-    elif "." not in m:
-        s = "" if m[1:] == "1" else "s"
-        return f"{INFLECT_ENGINE.number_to_words(m[1:])} {bill}{s}"
-    b, c = m[1:].split(".")
-    s = "" if b == "1" else "s"
-    c = int(c.ljust(2, "0"))
-    coins = (
-        f"cent{'' if c == 1 else 's'}"
-        if m[0] == "$"
-        else ("penny" if c == 1 else "pence")
-    )
-    return f"{INFLECT_ENGINE.number_to_words(b)} {bill}{s} and {INFLECT_ENGINE.number_to_words(c)} {coins}"
 
+    bill = "dollar" if m.group(2) == "$" else "pound"
+    coin = "cent" if m.group(2) == "$" else "pence"
+    number = m.group(3)
+
+    multiplier = m.group(4)
+    try:
+        number = float(number)
+    except:
+        return m.group()
+    
+    if m.group(1) == "-":
+        number *= -1
+
+    if number % 1 == 0 or multiplier != "":
+        text_number = f"{INFLECT_ENGINE.number_to_words(conditional_int(number))}{multiplier} {INFLECT_ENGINE.plural(bill, count=number)}"
+    else:
+        sub_number = int(str(number).split(".")[-1].ljust(2, "0"))
+
+        text_number = f"{INFLECT_ENGINE.number_to_words(int(round(number)))} {INFLECT_ENGINE.plural(bill, count=number)} and {INFLECT_ENGINE.number_to_words(sub_number)} {INFLECT_ENGINE.plural(coin, count=sub_number)}"
+
+    return text_number
 
 def handle_decimal(num: re.Match[str]) -> str:
     """Convert decimal numbers to spoken form"""
@@ -297,7 +309,7 @@ def normalize_text(text: str,normalization_options: NormalizationOptions) -> str
     text = re.sub(r"(?<=\d),(?=\d)", "", text)
     
     text = re.sub(
-        r"(?i)[$£]\d+(?:\.\d+)?(?: hundred| thousand| (?:[bm]|tr)illion)*\b|[$£]\d+\.\d\d?\b",
+        r"(?i)(-?)([$£])(\d+(?:\.\d+)?)((?: hundred| thousand| (?:[bm]|tr|quadr)illion)*)\b",
         handle_money,
         text,
     )

api/src/services/text_processing/text_processor.py

@@ -134,6 +134,7 @@ async def smart_split(
 
     # Normalize text
     if settings.advanced_text_normalization and normalization_options.normalize:
+        print(lang_code)
         if lang_code in ["a","b","en-us","en-gb"]:
             text = CUSTOM_PHONEMES.sub(lambda s: handle_custom_phonemes(s, custom_phoneme_list), text)
             text=normalize_text(text,normalization_options)

api/src/services/tts_service.py

@@ -258,7 +258,7 @@ class TTSService:
             logger.info(f"Using lang_code '{pipeline_lang_code}' for voice '{voice_name}' in audio stream")
 
             # Process text in chunks with smart splitting
-            async for chunk_text, tokens in smart_split(text, lang_code=lang_code, normalization_options=normalization_options):
+            async for chunk_text, tokens in smart_split(text, lang_code=pipeline_lang_code, normalization_options=normalization_options):
                 try:
                     # Process audio for chunk
                     async for chunk_data in self._process_chunk(

api/tests/test_kokoro_v1.py

@@ -23,19 +23,18 @@ def test_initial_state(kokoro_backend):
 
 
 @patch("torch.cuda.is_available", return_value=True)
-@patch("torch.cuda.memory_allocated")
+@patch("torch.cuda.memory_allocated", return_value=5e9)
 def test_memory_management(mock_memory, mock_cuda, kokoro_backend):
     """Test GPU memory management functions."""
-    # Mock GPU memory usage
-    mock_memory.return_value = 5e9  # 5GB
+    # Patch backend so it thinks we have cuda
+    with patch.object(kokoro_backend, "_device", "cuda"):
+        # Test memory check
+        with patch("api.src.inference.kokoro_v1.model_config") as mock_config:
+            mock_config.pytorch_gpu.memory_threshold = 4
+            assert kokoro_backend._check_memory() == True
 
-    # Test memory check
-    with patch("api.src.inference.kokoro_v1.model_config") as mock_config:
-        mock_config.pytorch_gpu.memory_threshold = 4
-        assert kokoro_backend._check_memory() == True
-
-        mock_config.pytorch_gpu.memory_threshold = 6
-        assert kokoro_backend._check_memory() == False
+            mock_config.pytorch_gpu.memory_threshold = 6
+            assert kokoro_backend._check_memory() == False
 
 
 @patch("torch.cuda.empty_cache")

api/tests/test_normalizer.py

@@ -83,7 +83,12 @@ def test_url_email_addresses():
         == "Send to test dot user at site dot com"
     )
 
-
+def test_money():
+    """Test that money text is normalized correctly"""
+    assert normalize_text("He lost $5.3 thousand.",normalization_options=NormalizationOptions()) == "He lost five point three thousand dollars."
+    assert normalize_text("To put it weirdly -$6.9 million",normalization_options=NormalizationOptions()) == "To put it weirdly minus six point nine million dollars"
+    assert normalize_text("It costs $50.3.",normalization_options=NormalizationOptions()) == "It costs fifty dollars and thirty cents."
+ 
 def test_non_url_text():
     """Test that non-URL text is unaffected"""
     assert normalize_text("This is not.a.url text",normalization_options=NormalizationOptions()) == "This is not-a-url text"

api/tests/test_text_processor.py

@@ -34,7 +34,7 @@ def test_process_text_chunk_phonemes():
 def test_get_sentence_info():
     """Test sentence splitting and info extraction."""
     text = "This is sentence one. This is sentence two! What about three?"
-    results = get_sentence_info(text)
+    results = get_sentence_info(text, {})
 
     assert len(results) == 3
     for sentence, tokens, count in results:
@@ -44,6 +44,19 @@ def test_get_sentence_info():
         assert count == len(tokens)
         assert count > 0
 
+def test_get_sentence_info_phenomoes():
+    """Test sentence splitting and info extraction."""
+    text = "This is sentence one. This is </|custom_phonemes_0|/> two! What about three?"
+    results = get_sentence_info(text, {"</|custom_phonemes_0|/>": r"sˈɛntᵊns"})
+
+    assert len(results) == 3
+    assert "sˈɛntᵊns" in results[1][0]
+    for sentence, tokens, count in results:
+        assert isinstance(sentence, str)
+        assert isinstance(tokens, list)
+        assert isinstance(count, int)
+        assert count == len(tokens)
+        assert count > 0
 
 @pytest.mark.asyncio
 async def test_smart_split_short_text():

dev/Test money.py

@@ -0,0 +1,26 @@
+import requests
+import base64
+import json
+
+text="""the administration has offered up a platter of repression for more than a year and is still slated to lose $400 million.
+
+Columbia is the largest private landowner in New York City and boasts an endowment of $14.8 billion;"""
+
+
+Type="wav"
+
+response = requests.post(
+    "http://localhost:8880/v1/audio/speech",
+    json={
+        "model": "kokoro",
+        "input": text,
+        "voice": "af_heart+af_sky",
+        "speed": 1.0,
+        "response_format": Type,
+        "stream": False,
+    },
+    stream=True
+)
+
+with open(f"outputnostreammoney.{Type}", "wb") as f:
+    f.write(response.content)

dev/Test num.py

@@ -0,0 +1,45 @@
+from text_to_num import text2num
+import re
+import inflect
+from torch import mul
+
+INFLECT_ENGINE = inflect.engine()
+
+
+def conditional_int(number: float, threshold: float = 0.00001):
+    if abs(round(number) - number) < threshold:
+        return int(round(number))
+    return number
+
+def handle_money(m: re.Match[str]) -> str:
+    """Convert money expressions to spoken form"""
+
+    bill = "dollar" if m.group(2) == "$" else "pound"
+    coin = "cent" if m.group(2) == "$" else "pence"
+    number = m.group(3)
+
+    multiplier = m.group(4)
+    try:
+        number = float(number)
+    except:
+        return m.group()
+    
+    if m.group(1) == "-":
+        number *= -1
+
+    if number % 1 == 0 or multiplier != "":
+        text_number = f"{INFLECT_ENGINE.number_to_words(conditional_int(number))}{multiplier} {INFLECT_ENGINE.plural(bill, count=number)}"
+    else:
+        sub_number = int(str(number).split(".")[-1].ljust(2, "0"))
+
+        text_number = f"{INFLECT_ENGINE.number_to_words(int(round(number)))} {INFLECT_ENGINE.plural(bill, count=number)} and {INFLECT_ENGINE.number_to_words(sub_number)} {INFLECT_ENGINE.plural(coin, count=sub_number)}"
+
+    return text_number
+
+
+text = re.sub(
+    r"(?i)(-?)([$£])(\d+(?:\.\d+)?)((?: hundred| thousand| (?:[bm]|tr|quadr)illion)*)\b",
+    handle_money,
+    "he administration has offered up a platter of repression for more than a year and is still slated to lose -$5.3 billion",
+)
+print(text)

pyproject.toml

@@ -38,6 +38,7 @@ dependencies = [
     "inflect>=7.5.0",
     "phonemizer-fork>=3.3.2",
     "av>=14.2.0",
+    "text2num>=2.5.1",
 ]
 
 [project.optional-dependencies]