From 9b4e9cc8a0311e5243d69b73ed073e7ea441982e Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 27 三月 2024 16:05:29 +0800
Subject: [PATCH] train update
---
funasr/models/uniasr/template.yaml | 54 +++++++++++++++++++++++++++++++++++++++++-------------
1 files changed, 41 insertions(+), 13 deletions(-)
diff --git a/funasr/models/uniasr/template.yaml b/funasr/models/uniasr/template.yaml
index f4815c1..e72a2d5 100644
--- a/funasr/models/uniasr/template.yaml
+++ b/funasr/models/uniasr/template.yaml
@@ -18,6 +18,7 @@
decoder_attention_chunk_type2: chunk
loss_weight_model1: 0.5
+
# encoder
encoder: SANMEncoderChunkOpt
encoder_conf:
@@ -34,11 +35,21 @@
kernel_size: 11
sanm_shfit: 0
selfattention_layer_type: sanm
- chunk_size: [20, 60]
- stride: [10, 40]
- pad_left: [5, 10]
- encoder_att_look_back_factor: [0, 0]
- decoder_att_look_back_factor: [0, 0]
+ chunk_size:
+ - 20
+ - 60
+ stride:
+ - 10
+ - 40
+ pad_left:
+ - 5
+ - 10
+ encoder_att_look_back_factor:
+ - 0
+ - 0
+ decoder_att_look_back_factor:
+ - 0
+ - 0
# decoder
decoder: FsmnDecoderSCAMAOpt
@@ -55,6 +66,7 @@
kernel_size: 11
concat_embeds: true
+# predictor
predictor: CifPredictorV2
predictor_conf:
idim: 320
@@ -62,6 +74,8 @@
l_order: 1
r_order: 1
+
+# encoder2
encoder2: SANMEncoderChunkOpt
encoder2_conf:
output_size: 320
@@ -77,12 +91,23 @@
kernel_size: 21
sanm_shfit: 0
selfattention_layer_type: sanm
- chunk_size: [45, 70]
- stride: [35, 50]
- pad_left: [5, 10]
- encoder_att_look_back_factor: [0, 0]
- decoder_att_look_back_factor: [0, 0]
+ chunk_size:
+ - 45
+ - 70
+ stride:
+ - 35
+ - 50
+ pad_left:
+ - 5
+ - 10
+ encoder_att_look_back_factor:
+ - 0
+ - 0
+ decoder_att_look_back_factor:
+ - 0
+ - 0
+# decoder
decoder2: FsmnDecoderSCAMAOpt
decoder2_conf:
attention_dim: 320
@@ -108,10 +133,12 @@
stride_conv_conf:
kernel_size: 2
stride: 2
- pad: [0, 1]
+ pad:
+ - 0
+ - 1
# frontend related
-frontend: WavFrontendOnline
+frontend: WavFrontend
frontend_conf:
fs: 16000
window: hamming
@@ -120,6 +147,7 @@
frame_shift: 10
lfr_m: 7
lfr_n: 6
+ dither: 0.0
specaug: SpecAugLFR
specaug_conf:
@@ -156,7 +184,7 @@
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
- batch_sampler: DynamicBatchLocalShuffleSampler
+ batch_sampler: BatchSampler
batch_type: example # example or length
batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
--
Gitblit v1.9.1