fix hyparam file
Browse files- hyperparams.yaml +8 -8
hyperparams.yaml
CHANGED
@@ -1,18 +1,18 @@
|
|
1 |
# ############################################################################
|
2 |
-
# Model: E2E ASR with Transformer
|
3 |
-
# Encoder:
|
4 |
# Decoder: Transformer Decoder + (CTC/ATT joint) beamsearch
|
5 |
# Tokens: BPE with unigram
|
6 |
# losses: CTC + KLdiv (Label Smoothing loss)
|
7 |
# Training: AISHELL-1
|
8 |
-
# Authors: Jianyuan Zhong, Titouan Parcollet
|
9 |
# ############################################################################
|
10 |
|
11 |
# Feature parameters
|
12 |
sample_rate: 16000
|
13 |
n_fft: 400
|
14 |
n_mels: 80
|
15 |
-
wav2vec2_hub: facebook/wav2vec2-large-
|
16 |
|
17 |
####################### Model parameters ###########################
|
18 |
# Transformer
|
@@ -36,7 +36,7 @@ unk_index: 0
|
|
36 |
|
37 |
# Decoding parameters
|
38 |
min_decode_ratio: 0.0
|
39 |
-
max_decode_ratio: 1.0
|
40 |
valid_search_interval: 10
|
41 |
valid_beam_size: 10
|
42 |
test_beam_size: 10
|
@@ -49,7 +49,7 @@ wav2vec2: !new:speechbrain.lobes.models.huggingface_wav2vec.HuggingFaceWav2Vec2
|
|
49 |
output_norm: True
|
50 |
freeze: True
|
51 |
pretrain: False # Pretraining is managed by the SpeechBrain pre-trainer.
|
52 |
-
save_path:
|
53 |
|
54 |
Transformer: !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
|
55 |
input_size: 1024
|
@@ -108,6 +108,6 @@ log_softmax: !new:torch.nn.LogSoftmax
|
|
108 |
|
109 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
110 |
loadables:
|
111 |
-
|
112 |
-
model: !ref <
|
113 |
tokenizer: !ref <tokenizer>
|
|
|
1 |
# ############################################################################
|
2 |
+
# Model: E2E ASR with Transformer + wav2vec2
|
3 |
+
# Encoder: wav2vec Encoder
|
4 |
# Decoder: Transformer Decoder + (CTC/ATT joint) beamsearch
|
5 |
# Tokens: BPE with unigram
|
6 |
# losses: CTC + KLdiv (Label Smoothing loss)
|
7 |
# Training: AISHELL-1
|
8 |
+
# Authors: Jianyuan Zhong, Titouan Parcollet, Mirco Ravanelli
|
9 |
# ############################################################################
|
10 |
|
11 |
# Feature parameters
|
12 |
sample_rate: 16000
|
13 |
n_fft: 400
|
14 |
n_mels: 80
|
15 |
+
wav2vec2_hub: facebook/wav2vec2-large-100k-voxpopuli
|
16 |
|
17 |
####################### Model parameters ###########################
|
18 |
# Transformer
|
|
|
36 |
|
37 |
# Decoding parameters
|
38 |
min_decode_ratio: 0.0
|
39 |
+
max_decode_ratio: 1.0
|
40 |
valid_search_interval: 10
|
41 |
valid_beam_size: 10
|
42 |
test_beam_size: 10
|
|
|
49 |
output_norm: True
|
50 |
freeze: True
|
51 |
pretrain: False # Pretraining is managed by the SpeechBrain pre-trainer.
|
52 |
+
save_path: model_checkpoints
|
53 |
|
54 |
Transformer: !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
|
55 |
input_size: 1024
|
|
|
108 |
|
109 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
110 |
loadables:
|
111 |
+
wav2vec2: !ref <wav2vec2>
|
112 |
+
model: !ref <asr_model>
|
113 |
tokenizer: !ref <tokenizer>
|