1

我在从头开始训练模型后尝试解码 wav 文件，我完成了训练和测试阶段没有错误，我得到了 WER & CER 和损失值。注意：我已经用这个命令完成了解码，没有错误，但是当我用新模型测试时它出现了。

native_client/deepspeech --model /home/xyz/DeepSpeech/data/exprt_dir/output_graph.pb --scorer /home/xyz/DeepSpeech/data/lm/lm.scorer --audio data/test5.wav --beam_width 9000 > data/decoding.txt
TensorFlow: v2.2.0-15-g518c1d0
DeepSpeech: v0.9.0-alpha.3-0-g78ae08c
Warning: reading entire model file into memory. Transform model file into an  mmapped graph to reduce heap usage.
2020-09-22 12:15:00.293543: I tensorflow/core/platform/cpu_feature_guard.cc:143]  Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
terminate called after throwing an instance of 'std::length_error'
what():  vector::_M_default_append
Abandon (core dumped)`

我尝试了没有记分器、管道和梁的命令以及来自火车数据集的文件，但错误仍然存​​在。

native_client/deepspeech --model /home/xyz/DeepSpeech/data/exprt_dir/output_graph.pb --audio data/decoding_online/test5.wav
TensorFlow: v2.2.0-15-g518c1d0
DeepSpeech: v0.9.0-alpha.3-0-g78ae08c
Warning: reading entire model file into memory. Transform model file into an mmapped graph to reduce heap usage.
2020-09-22 13:56:17.511266: I tensorflow/core/platform/cpu_feature_guard.cc:143]       Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
terminate called after throwing an instance of 'std::length_error'
what():  vector::_M_default_append
Abandon (core dumped)

您可以在下面找到我拥有的软件包版本（运行 pip list 显示的不仅仅是这些软件包，但我只保留了重要的软件包）

(deepspeech-venv) (base) root@xyz:/home/xyz/DeepSpeech# pip  list
Package              Version      Location                            
-------------------- ------------ ------------------------------------
deepspeech-training  0.9.0a3      /home/xyz/DeepSpeech/training
ds-ctcdecoder        0.9.0a3      
tensorboard          1.15.0       
tensorflow-estimator 1.15.1       
tensorflow-gpu       1.15.2
4

1 回答 1

0

只是为了将来参考，这个问题在Discourse上进行了详细讨论。

一般来说，如果您没有积极开发 DeepSpeech，请不要使用 master 分支，因为它有时会中断。

于 2020-11-04T08:54:36.897 回答