upload better model

.gitattributes

@@ -29,3 +29,5 @@ asr.ckpt filter=lfs diff=lfs merge=lfs -text
 lm.ckpt filter=lfs diff=lfs merge=lfs -text
 normalizer.ckpt filter=lfs diff=lfs merge=lfs -text
 tokenizer.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ckpt. filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text

README.md

@@ -30,7 +30,7 @@ The performance of the model is the following:
 
 | Release  | eval clean CER | eval other CER |    GPUs     |
 | :------: | :------------: | :------------: | :---------: |
-| 09-05-21 |     7.48%      |     8.38%      | 6xA100 80GB |
+| 22-07-23 |     7.33%      |     7.99%      | 6xA100 80GB |
 
 ## Pipeline description
 

asr.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:893a5fb84a67315a954d7645fd3b5f96cee806531f538e0073f6dcdf17dcf7c3
-size 183510489
+oid sha256:8e718dc29b403dfaa8d2604c43c3666be3fa99e958b77e3c6ff387e94d4a174c
+size 184546287

hyperparams.yaml

@@ -22,14 +22,14 @@ nhead: 4
 num_encoder_layers: 12
 num_decoder_layers: 6
 d_ffn: 2048
-transformer_dropout: 0.0
+transformer_dropout: 0.1
 activation: !name:torch.nn.GELU
 output_neurons: 5000
 vocab_size: 5000
 
 # Outputs
 blank_index: 0
-label_smoothing: 0.1
+label_smoothing: 0.0
 pad_index: 0
 bos_index: 1
 eos_index: 2
@@ -40,9 +40,9 @@ min_decode_ratio: 0.0
 max_decode_ratio: 1.0
 valid_search_interval: 10
 valid_beam_size: 10
-test_beam_size: 60
-lm_weight: 0.20
-ctc_weight_decode: 0.40
+test_beam_size: 66
+lm_weight: 0.60
+ctc_weight_decode: 0.50
 
 ############################## models ################################
 
@@ -51,12 +51,12 @@ normalizer: !new:speechbrain.processing.features.InputNormalization
 
 CNN: !new:speechbrain.lobes.models.convolution.ConvolutionFrontEnd
     input_shape: (8, 10, 80)
-    num_blocks: 2
+    num_blocks: 3
     num_layers_per_block: 1
-    out_channels: (64, 32)
-    kernel_sizes: (3, 3)
-    strides: (2, 2)
-    residuals: (False, False)
+    out_channels: (64, 64, 64)
+    kernel_sizes: (5, 5, 1)
+    strides: (2, 2, 1)
+    residuals: (False, False, True)
 
 Transformer: !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
     input_size: 640
@@ -106,8 +106,8 @@ decoder: !new:speechbrain.decoders.S2STransformerBeamSearch
     ctc_weight: !ref <ctc_weight_decode>
     lm_weight: !ref <lm_weight>
     lm_modules: !ref <lm_model>
-    temperature: 1.25
-    temperature_lm: 1.25
+    temperature: 1.15
+    temperature_lm: 1.15
     using_eos_threshold: False
     length_normalization: True
 

lm.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee4a5a5d9ce11e24dcea93f24a241528b9b376798be6478c70fb279736515110
-size 381074814
+oid sha256:7f0b49d5e1f9894c0c9f2ec21c8658da8e1a07f509b807e8624450ba19ea667c
+size 381072461

normalizer.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4866d96b29f5c97526c7469aa6f58cd50aeb9865b457daf599f0f42e5827be9
-size 1783
+oid sha256:1da2ced935d955c014177591249e5db497d0c5dc7143e64378da0cb5590fe77a
+size 1703

tokenizer.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e095c023a42b6bd25352512597a245db9bf9126ce6bf64082bd41d0a196b220
-size 313899
+oid sha256:d419e55734c26df6c5690671be2b887a7db389c1a7f63286111ce737508c6569
+size 313900