Fix JSON serialization issue with numpy.float32 values in voicevox se…

…rver

This commit introduces a new utility function `convert_to_float` that recursively converts numpy.float32 values to Python's native float type. This change is necessary because numpy.float32 values are not serializable by Python's json library, which causes issues when these values need to be sent as JSON responses from the voicevox server.

The `convert_to_float` function is now applied to `accent_phrases` after they are generated by the engine, ensuring that all numeric values are in a format compatible with JSON serialization standards.

This update ensures that the server can handle serialization of voice properties without encountering errors due to data type compatibility issues with JSON.

3rdparty/voicevox/node_scripts/server.py

@@ -19,6 +19,7 @@
 from typing import Optional
 import zipfile
 
+import numpy as np
 from fastapi import FastAPI
 from fastapi import HTTPException
 from fastapi.middleware.cors import CORSMiddleware
@@ -53,6 +54,7 @@
 from voicevox_engine.model import Speaker
 from voicevox_engine.model import SpeakerInfo
 from voicevox_engine.model import SupportedDevicesInfo
+from voicevox_engine.model import Mora
 from voicevox_engine.morphing import \
     synthesis_morphing_parameter as _synthesis_morphing_parameter
 from voicevox_engine.morphing import synthesis_morphing
@@ -66,6 +68,21 @@
 from voicevox_engine.utility import engine_root
 
 
+def convert_to_float(obj):
+    """
+    Recursively converts numpy.float32 values to Python float in an object's attributes.
+    """
+    if isinstance(obj, list):
+        return [convert_to_float(item) for item in obj]
+    elif isinstance(obj, AccentPhrase) or isinstance(obj, Mora):
+        for attr, value in obj.__dict__.items():
+            setattr(obj, attr, convert_to_float(value))
+        return obj
+    elif isinstance(obj, np.float32):
+        return float(obj)
+    else:
+        return obj
+
 def b64encode_str(s):
     return base64.b64encode(s).decode("utf-8")
 
@@ -129,6 +146,7 @@ def audio_query(text: str, speaker: int, core_version: Optional[str] = None):
         """
         engine = get_engine(core_version)
         accent_phrases = engine.create_accent_phrases(text, speaker_id=speaker)
+        accent_phrases = [convert_to_float(ac) for ac in accent_phrases]
         return AudioQuery(
             accent_phrases=accent_phrases,
             speedScale=1,