Tensorflow Graph已固化无法修改报错，手动重置无效如何解决？

使用inaSpeechSegmenter进行音频分段时触发TensorFlow图已固化的错误，代码如下：

vocals_path = "output/song.wav"
audio_path = vocals_path
from inaSpeechSegmenter import Segmenter
import tensorflow as tf

# Initialize the segmenter
seg = Segmenter()

# Analyze the audio file
segments = seg(vocals_path)

# Print the segments
for segment in segments:
    print(segment)

File "/root/code/aimusic/essgender.py", line 138, in <module>
    seg = Segmenter()
          ^^^^^^^^^^^
  File "/root/code/aimusic/venv/lib/python3.11/site-packages/inaSpeechSegmenter/segmenter.py", line 241, in __init__
    self.vad = SpeechMusicNoise(batch_size)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/code/aimusic/venv/lib/python3.11/site-packages/inaSpeechSegmenter/segmenter.py", line 131, in __init__
    self.nn = keras.models.load_model(model_path, compile=False)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/code/aimusic/venv/lib/python3.11/site-packages/keras/saving/saving_api.py", line 212, in load_model
    return legacy_sm_saving_lib.load_model(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/code/aimusic/venv/lib/python3.11/site-packages/keras/utils/traceback_utils.py", line 70, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/root/code/aimusic/venv/lib/python3.11/site-packages/tensorflow/python/framework/ops.py", line 3375, in _check_not_finalized
    raise RuntimeError("Graph is finalized and cannot be modified.")
RuntimeError: Graph is finalized and cannot be modified.

• 手动重置默认图：

tf.compat.v1.reset_default_graph()

• 重置图并创建交互式会话：

from tensorflow.python.framework import ops
ops.reset_default_graph()
sess = tf.InteractiveSession()

• 调整导入顺序：先导入TensorFlow，再导入inaSpeechSegmenter，避免图被提前初始化：

import tensorflow as tf
from inaSpeechSegmenter import Segmenter

vocals_path = "output/song.wav"
audio_path = vocals_path

seg = Segmenter()
segments = seg(vocals_path)

for segment in segments:
    print(segment)

• 确保版本兼容：inaSpeechSegmenter与TensorFlow版本存在适配关系，若当前版本不兼容，降级到稳定组合：

pip uninstall tensorflow inaSpeechSegmenter
pip install tensorflow==2.7 inaSpeechSegmenter==0.7.5

• 禁用TF1.x兼容模式：如果脚本中存在tf.compat.v1相关调用，全部移除，使用TensorFlow 2.x原生的eager execution模式（默认开启），避免手动创建会话或操作图。
• 检查前置代码：确认脚本其他部分没有提前初始化TensorFlow会话、冻结图的操作，若有则将Segmenter的初始化移至这些操作之前。

内容的提问来源于stack exchange，提问作者user2950593