piper-tts-plus 20250622.114312__py3-none-any.whl → 20250624.105509__py3-none-any.whl

piper/voice.py

@@ -7,8 +7,32 @@ from typing import Any, Dict, Iterable, List, Optional, Tuple, Union
 
 import numpy as np
 import onnxruntime
-import pyopenjtalk
-from piper_phonemize import phonemize_codepoints, phonemize_espeak, tashkeel_run
+
+# Try to import piper_phonemize, but make it optional
+try:
+    from piper_phonemize import phonemize_codepoints, phonemize_espeak, tashkeel_run
+    HAS_PIPER_PHONEMIZE = True
+except ImportError:
+    HAS_PIPER_PHONEMIZE = False
+    # Provide fallback implementations
+    def phonemize_codepoints(text, lang=None):
+        # Simple fallback: return text as list of characters
+        return list(text)
+    
+    def phonemize_espeak(text, voice=None):
+        # Simple fallback: return text as list of characters
+        return list(text)
+    
+    def tashkeel_run(text):
+        # Simple fallback: return original text
+        return text
+
+# Try to import pyopenjtalk, but make it optional
+try:
+    import pyopenjtalk
+    HAS_PYOPENJTALK = True
+except ImportError:
+    HAS_PYOPENJTALK = False
 
 from .config import PhonemeType, PiperConfig
 from .const import BOS, EOS, PAD
@@ -16,6 +40,33 @@ from .util import audio_float_to_int16
 
 _LOGGER = logging.getLogger(__name__)
 
+# Multi-character phoneme to PUA character mapping for Japanese
+# This must match the C++ side and Python training side
+MULTI_CHAR_TO_PUA = {
+    "a:": "\ue000",
+    "i:": "\ue001",
+    "u:": "\ue002",
+    "e:": "\ue003",
+    "o:": "\ue004",
+    "cl": "\ue005",
+    "ky": "\ue006",
+    "kw": "\ue007",
+    "gy": "\ue008",
+    "gw": "\ue009",
+    "ty": "\ue00a",
+    "dy": "\ue00b",
+    "py": "\ue00c",
+    "by": "\ue00d",
+    "ch": "\ue00e",
+    "ts": "\ue00f",
+    "sh": "\ue010",
+    "zy": "\ue011",
+    "hy": "\ue012",
+    "ny": "\ue013",
+    "my": "\ue014",
+    "ry": "\ue015",
+}
+
 
 @dataclass
 class PiperVoice:
@@ -81,16 +132,47 @@ class PiperVoice:
                 phonemes = pyopenjtalk.g2p(text, kana=False).split()
 
                 converted = []
+                # Add BOS marker
+                converted.append("^")
+                
                 for ph in phonemes:
                     if ph == "pau":
                         converted.append("_")
                         continue
+                    
+                    if ph == "sil":
+                        # Skip sil in the middle, it will be added as EOS
+                        continue
 
                     # Devoiced vowels come back as upper-case (A,I,U,E,O)
+                    # But NOT 'N' which is a special phoneme
                     if ph in {"A", "I", "U", "E", "O"}:
                         ph = ph.lower()
-
-                    converted.append(ph)
+                    
+                    # Check if this is a multi-character phoneme that needs PUA mapping
+                    if ph in MULTI_CHAR_TO_PUA:
+                        converted.append(MULTI_CHAR_TO_PUA[ph])
+                    else:
+                        converted.append(ph)
+                
+                # Add EOS marker
+                converted.append("$")
+                
+                # Log readable phonemes if debug logging is enabled
+                if _LOGGER.isEnabledFor(logging.DEBUG):
+                    readable_phonemes = []
+                    for ph in converted:
+                        if len(ph) == 1 and ord(ph) >= 0xE000 and ord(ph) <= 0xF8FF:
+                            # Find the original multi-char phoneme
+                            for orig, pua in MULTI_CHAR_TO_PUA.items():
+                                if pua == ph:
+                                    readable_phonemes.append(orig)
+                                    break
+                            else:
+                                readable_phonemes.append(ph)
+                        else:
+                            readable_phonemes.append(ph)
+                    _LOGGER.debug("Phonemized '%s' to: %s", text, ' '.join(readable_phonemes))
 
                 return [converted]
 

{piper_tts_plus-20250622.114312.dist-info → piper_tts_plus-20250624.105509.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: piper-tts-plus
-Version: 20250622.114312
+Version: 20250624.105509
 Summary: A fast, local neural text to speech system that sounds great and is optimized for the Raspberry Pi 4.
 Home-page: https://github.com/ayutaz/piper-plus
 Author: yousan

{piper_tts_plus-20250622.114312.dist-info → piper_tts_plus-20250624.105509.dist-info}/RECORD

@@ -6,10 +6,10 @@ piper/download.py,sha256=zyF2oyvuZiQr1HiAIBspiQByAcGOkXJfTj_KOecImxU,4604
 piper/file_hash.py,sha256=HMuwrgEIg-bCOXHG0wE3vtjrqGD7QaA_UNfvBMXeUcY,1107
 piper/http_server.py,sha256=12B9PJCY4UN_fcoH-gunxq7o3obj7NEfpXgo9tESKR4,4065
 piper/util.py,sha256=QQnvx_HZGUk9gHCsqifR1ob7-2QRfN9jZMqF0poQS8k,397
-piper/voice.py,sha256=u2tVneH4_jgrcM7QQu-1qRA24_VKlI3ZjrxV5gH7_Yc,7363
+piper/voice.py,sha256=jCCoipLyHD44XYA1esE4HLftdPr13nB9i5Zl0MXqKLY,10172
 piper/voices.json,sha256=pusxO-rZHvmWXkbRpp05lQCwYns7oKXAU6SIH3wFkUQ,130859
-piper_tts_plus-20250622.114312.dist-info/METADATA,sha256=9FH3CUkkfQ7yaOP9g1_0_ITGP0c50fCaQqu9ShUez8M,1201
-piper_tts_plus-20250622.114312.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-piper_tts_plus-20250622.114312.dist-info/entry_points.txt,sha256=R-zJAXMPM47DHJDk5K0toz7iOATx38RV98fkowMxj0o,46
-piper_tts_plus-20250622.114312.dist-info/top_level.txt,sha256=3FjLp04vySCS6YYps6jbChyOvij45fGR-MHPpBFUIlw,6
-piper_tts_plus-20250622.114312.dist-info/RECORD,,
+piper_tts_plus-20250624.105509.dist-info/METADATA,sha256=NhhN-9kpge9xgfb6mDFFv-D5tfunCnV0qk8PZMRePq0,1201
+piper_tts_plus-20250624.105509.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+piper_tts_plus-20250624.105509.dist-info/entry_points.txt,sha256=R-zJAXMPM47DHJDk5K0toz7iOATx38RV98fkowMxj0o,46
+piper_tts_plus-20250624.105509.dist-info/top_level.txt,sha256=3FjLp04vySCS6YYps6jbChyOvij45fGR-MHPpBFUIlw,6
+piper_tts_plus-20250624.105509.dist-info/RECORD,,