funasr1.0.2

游雁

2024-01-25 369382050bf71c249944545f009a29a8632fdda5

 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: