声音分类ESC50 Example模型部署出错！TypeError: only size-1 arrays can be converted to Python scalars

[policemanxxx]

【过程】
模型训练->模型预测->动转静->模型部署

【错误描述】
通过命令 ./run.sh 4 gpu ./export /audio/dog.wav 进行第4步骤。
无法正确执行，尝试各种方法无果，请求协助！

【环境变量】
paddleaudio 1.0.2
paddlepaddle-gpu 2.5.0
paddlespeech 0.0.0

【具体报错如下】
I0809 15:22:10.513600 24904 analysis_predictor.cc:1660] ======= optimize end =======
I0809 15:22:10.513777 24904 naive_executor.cc:164] --- skip [feed], feed -> x
I0809 15:22:10.514379 24904 naive_executor.cc:164] --- skip [linear_4.tmp_1], fetch -> fetch
W0809 15:22:10.516772 24904 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.4, Runtime API Version: 11.2
W0809 15:22:10.518486 24904 gpu_resources.cc:149] device: 0, cuDNN Version: 8.1.
Traceback (most recent call last):
File "/home/workstation/chenwh/PaddleSpeech/paddlespeech/cls/exps/panns/deploy/predict.py", line 144, in
results = predictor.predict(wavs)
File "/home/workstation/chenwh/PaddleSpeech/paddlespeech/cls/exps/panns/deploy/predict.py", line 120, in predict
feats = extract_features(wavs)
File "/home/workstation/chenwh/PaddleSpeech/paddlespeech/cls/exps/panns/deploy/predict.py", line 58, in extract_features
feat = MelSpectrogram(waveforms[i], sr, **kwargs).transpose()
File "/root/anaconda3/envs/cwh-test/lib/python3.8/site-packages/paddle/audio/features/layers.py", line 181, in init
self.fbank_matrix = compute_fbank_matrix(
File "/root/anaconda3/envs/cwh-test/lib/python3.8/site-packages/paddle/audio/functional/functional.py", line 227, in compute_fbank_matrix
fftfreqs = fft_frequencies(sr=sr, n_fft=n_fft, dtype=dtype)
File "/root/anaconda3/envs/cwh-test/lib/python3.8/site-packages/paddle/audio/functional/functional.py", line 183, in fft_frequencies
return paddle.linspace(0, float(sr) / 2, int(1 + n_fft // 2), dtype=dtype)
TypeError: only size-1 arrays can be converted to Python scalars