blob: 51e30f4a5bbde92f0918e915992956b86cfa01d2 (
plain) (
blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
|
dataset:
download: True
dataset_root_path: "data/datasets" # files will be downloaded into this dir
language_name: "mls_german_opus"
limited_supervision: False # set to True if you want to use limited supervision
dataset_percentage: 1 # percentage of dataset to use (1.0 = 100%)
shuffle: True
model:
n_cnn_layers: 3
n_rnn_layers: 7
rnn_dim: 512
n_feats: 256 # number of mel features
stride: 2
dropout: 0.2 # recommended to be around 0.4 for smaller datasets, 0.1 for really large datasets
tokenizer:
tokenizer_path: "data/tokenizers/char_tokenizer_german.json"
decoder:
type: "greedy" # greedy, or lm (beam search)
lm: # config for lm decoder
language_model_path: "data" # path where model and supplementary files are stored
language: "german"
n_gram: 5 # n-gram size of the language model, 3 or 5
beam_size: 500
beam_threshold: 150
n_best: 1
lm_weight: 1
word_score: 1
training:
learning_rate: 0.0005
batch_size: 32 # recommended to maximum number that fits on the GPU (batch size of 32 fits on a 12GB GPU)
epochs: 100
eval_every_n: 5 # evaluate every n epochs
num_workers: 4 # number of workers for dataloader
checkpoints: # use "~" to disable saving/loading
model_load_path: ~ # path to load model from
model_save_path: "data/runs/01/epoch" # path to save model to
inference:
model_load_path: "data/epoch67" # path to load model from
|
