运行～/DeepSpeech/example/aishell/run_infer_golden.sh 报错。貌似以下两个参数的文件不匹配。
--model_path='models/aishell/params.tar.gz' 
--lang_model_path='models/lm/zh_giga.no_cna_cmn.prune01244.klm' \

报错信息如下。
[INFO 2018-04-10 10:57:46,684 model.py:243] begin to initialize the external scorer for decoding
terminate called after throwing an instance of 'lm::FormatLoadException'
what(): kenlm/lm/binary_format.cc:160 in void* lm::ngram::BinaryFormat::LoadBinary(std::size_t) threw FormatLoadException because `file_size != util::kBadSize && file_size < total_map'.
Binary file has size 1088258466 but the headers say it should be at least 2953349384

本次是在Ubuntu16.04 不使用docker下测试 DeepSpeech，且不使用GPU,运行参数如下。
CUDA_VISIBLE_DEVICES=0 
python -u infer.py 
--num_samples=10 
--trainer_count=1 
--beam_size=300 
--num_proc_bsearch=8 
--num_conv_layers=2 
--num_rnn_layers=3 
--rnn_layer_size=1024 
--alpha=2.6 
--beta=5.0 
--cutoff_prob=0.99 
--cutoff_top_n=40 
--use_gru=True 
--use_gpu=False 
--share_rnn_weights=False 
--infer_manifest='data/aishell/manifest.test' 
--mean_std_path='models/aishell/mean_std.npz' 
--vocab_path='models/aishell/vocab.txt' 
--model_path='models/aishell/params.tar.gz' 
--lang_model_path='models/lm/zh_giga.no_cna_cmn.prune01244.klm' 
--decoding_method='ctc_beam_search' 
--error_rate_type='cer' 
--specgram_type='linear'