Created by: harold-yh

系统环境：Ubuntu 16.04,使用的两块GTX 1080TI显卡，内存32G，SWAP：800G

参照aishell脚本训练自己的语音和标注数据，训练结束时报错： ................................................................................................... Pass: 18, Batch: 100, TrainCost: 70.923850 ............................................................................ ------- Time: 759 sec, Pass: 18, ValidationCost: 131.906596488 ................................................................................................... Pass: 19, Batch: 100, TrainCost: 70.486890 ............................................................................ ------- Time: 754 sec, Pass: 19, ValidationCost: 106.161569876

* Aborted at 1515694750 (unix time) try "date -d @1515694750" if you are using GNU date *

使用的参数如下： ----------- Configuration Arguments ----------- augment_conf_path: conf/augmentation.config batch_size: 8 dev_manifest: data/numbers/manifest.dev init_model_path: checkpoints/numbers/params.pass-19.tar.gz is_local: 1 learning_rate: 1e-06 max_duration: 27.0 mean_std_path: data/numbers/mean_std.npz min_duration: 0.0 num_conv_layers: 2 num_iter_print: 100 num_passes: 20 num_proc_data: 5 num_rnn_layers: 3 output_model_dir: ./checkpoints/numbers rnn_layer_size: 2048 share_rnn_weights: 1 shuffle_method: batch_shuffle_clipped specgram_type: linear test_off: 0 train_manifest: data/numbers/manifest.train trainer_count: 2 use_gpu: 1 use_gru: 1 use_sortagrad: 1 vocab_path: data/numbers/vocab.txt