|
|
|
|
# https://yaml.org/type/float.html
|
|
|
|
|
data:
|
|
|
|
|
train_manifest: data/manifest.train
|
|
|
|
|
dev_manifest: data/manifest.dev
|
|
|
|
|
test_manifest: data/manifest.test
|
|
|
|
|
min_input_len: 0.0
|
|
|
|
|
max_input_len: 27.0 # second
|
|
|
|
|
min_output_len: 0.0
|
|
|
|
|
max_output_len: .inf
|
|
|
|
|
min_output_input_ratio: 0.00
|
|
|
|
|
max_output_input_ratio: .inf
|
|
|
|
|
|
|
|
|
|
collator:
|
|
|
|
|
batch_size: 64 # one gpu
|
|
|
|
|
mean_std_filepath: data/mean_std.json
|
|
|
|
|
unit_type: char
|
|
|
|
|
vocab_filepath: data/vocab.txt
|
|
|
|
|
augmentation_config: conf/augmentation.json
|
|
|
|
|
random_seed: 0
|
|
|
|
|
spm_model_prefix:
|
|
|
|
|
specgram_type: linear #linear, mfcc, fbank
|
|
|
|
|
feat_dim:
|
|
|
|
|
delta_delta: False
|
|
|
|
|
stride_ms: 10.0
|
|
|
|
|
window_ms: 20.0
|
|
|
|
|
n_fft: None
|
|
|
|
|
max_freq: None
|
|
|
|
|
target_sample_rate: 16000
|
|
|
|
|
use_dB_normalization: True
|
|
|
|
|
target_dB: -20
|
|
|
|
|
dither: 1.0
|
|
|
|
|
keep_transcription_text: False
|
|
|
|
|
sortagrad: True
|
|
|
|
|
shuffle_method: batch_shuffle
|
|
|
|
|
num_workers: 0
|
|
|
|
|
|
|
|
|
|
model:
|
|
|
|
|
num_conv_layers: 2
|
|
|
|
|
num_rnn_layers: 3
|
|
|
|
|
rnn_layer_size: 1024
|
|
|
|
|
rnn_direction: forward # [forward, bidirect]
|
|
|
|
|
num_fc_layers: 1
|
|
|
|
|
fc_layers_size_list: 512,
|
|
|
|
|
use_gru: False
|
|
|
|
|
|
|
|
|
|
training:
|
|
|
|
|
n_epoch: 50
|
|
|
|
|
lr: 2e-3
|
|
|
|
|
lr_decay: 0.91 # 0.83
|
|
|
|
|
weight_decay: 1e-06
|
|
|
|
|
global_grad_clip: 3.0
|
|
|
|
|
log_interval: 100
|
|
|
|
|
checkpoint:
|
|
|
|
|
kbest_n: 50
|
|
|
|
|
latest_n: 5
|
|
|
|
|
|
|
|
|
|
decoding:
|
|
|
|
|
batch_size: 32
|
|
|
|
|
error_rate_type: cer
|
|
|
|
|
decoding_method: ctc_beam_search
|
|
|
|
|
lang_model_path: data/lm/zh_giga.no_cna_cmn.prune01244.klm
|
|
|
|
|
alpha: 1.9
|
|
|
|
|
beta: 5.0
|
|
|
|
|
beam_size: 300
|
|
|
|
|
cutoff_prob: 0.99
|
|
|
|
|
cutoff_top_n: 40
|
|
|
|
|
num_proc_bsearch: 10
|
