Tensor2tensor experiment with SpecAugment
Implementation of SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
echo "No specAugment"
# Set Paths
MODEL=transformer
HPARAMS=transformer_librispeech_v1
PROBLEM=librispeech_clean_small
DATA_DIR=data/no_spec
TMP_DIR=tmp
TRAIN_DIR=train/$PROBLEM
mkdir -p $DATA_DIR $TMP_DIR $TRAIN_DIR
# Generate data
t2t-datagen \
--data_dir=$DATA_DIR \
--tmp_dir=$TMP_DIR \
--problem=$PROBLEM
# Train
t2t-trainer \
--data_dir=$DATA_DIR \
--problem=$PROBLEM \
--model=$MODEL \
--hparams_set=$HPARAMS \
--output_dir=$TRAIN_DIR \
--train_steps=500000 \
--eval_steps=3 \
--local_eval_frequency=5000 \
--worker_gpu=4
echo "specAugment"
# Set Paths
PROBLEM=librispeech_specaugment
DATA_DIR=data/spec
TMP_DIR=tmp
TRAIN_DIR=train/$PROBLEM
USER_DIR=USER_DIR
mkdir -p $DATA_DIR $TMP_DIR $TRAIN_DIR
# Generate data
t2t-datagen \
--data_dir=$DATA_DIR \
--tmp_dir=$TMP_DIR \
--problem=$PROBLEM
# Train
t2t-trainer \
--t2t_usr_dir=$USER_DIR \
--data_dir=$DATA_DIR \
--problem=$PROBLEM \
--model=$MODEL \
--hparams_set=$HPARAMS \
--output_dir=$TRAIN_DIR \
--train_steps=500000 \
--eval_steps=3 \
--local_eval_frequency=5000 \
--worker_gpu=4