[Bug]: 在多次使用 pipeline('towhee/audio-embedding-vggish') 时发生内存泄漏

[wangdabee]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

embedding_pipeline = pipeline('towhee/audio-embedding-vggish')
outs1 = embedding_pipeline(wav_path)

在循环使用上述代码时，发生内存泄漏问题。达到20%内存

Expected Behavior

No response

Steps To Reproduce

No response

Environment

- Towhee version(e.g. v0.1.3 or 8b23a93):0.9.0
- OS(Ubuntu or CentOS):
- CPU/Memory:
- GPU:
- Others:

Anything else?

No response

[wangdabee]

此外，在循环使用代码时，会出现卡住的现象，即程序没有进行任何操作，也不报错。ctrl c 强制退出时 在 threading 相关代码处停止。

[junjiejiangjjj]

/assign @junjiejiangjjj

[junjiejiangjjj]

Upgrade Towhee to the latest version 1.1.1， and try it: https://towhee.io/towhee/audio-embedding

[wangdabee]

升级到1.1.1后 循环使用代码时，程序卡住的现象没有了，还是会有内存泄漏的现象，不过比0.9.0泄露的少一点

[junjiejiangjjj]

I have run this audio:https://towhee.io/towhee/audio-embedding/resolve/branch/main/test.wav 10,000 times. The memory usage has remained stable at around 2.9GB. I have not observed any memory leaks.

[wangdabee]

path1 = '1.wav'
path2 = '2.wav'

p = AutoPipes.pipeline('audio-embedding-vggish')

for i in range(1000):
res1 = p(path1)
res2 = p(path2)

res1 = res1.get()[0]
res2 = res2.get()[0]

运行以上代码，对源码的唯一修改是将device指定为cpu，内存占用大概12GB.
还有一个小问题，在运行完最后一次循环后，代码没有自动退出，一直卡住不动，需要手动ctrl c

[junjiejiangjjj]

I used this code snippet for testing, the memory was stable, and the problem could not be reproduced.

[wangdabee]

ok thank you for your patience

[wangdabee]

@junjiejiangjjj 您能说下您测试时使用的python版本吗

[junjiejiangjjj]

@junjiejiangjjj 您能说下您测试时使用的python版本吗

3.9.12

[wangdabee]

@junjiejiangjjj 解决了 我用的是python 3.7 发生的泄露