toliz
commented
5 years ago I opened the respective cpp file and I see the variables FLAGS_tokensdir and FLAGS_tokens are concatenated to form the dictionary path. But the problem is I don't know which file is assumed to be the dictionary and also I can't see the value of these flags to know where to place it. From this line I see that the flags should be saved in /tmp/Decoder.INFO, but my logger contains only the following:
Log line format: [IWEF]mmdd hh:mm:ss.uuuuuu threadid file:line] msg
I0828 10:05:39.377187 985 Decode.cpp:53] Parsing command line flags
I0828 10:05:39.377609 985 Decode.cpp:57] Reading flags from file decode_convlm_other.cfg
I0828 10:05:39.377884 985 Decode.cpp:82] [Network] Reading acoustic model from seq2seq_tds_librispeech/am/baseline_dev-other.bin
I0828 10:05:40.717425 985 Decode.cpp:86] [Network] Sequential [input -> (0) -> (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) -> output]
(0): View (-1 80 1 0)
(1): Conv2D (1->10, 5x21, 2,1, SAME,SAME, 1, 1) (with bias)
(2): ReLU
(3): Dropout (0.200000)
(4): LayerNorm ( axes : { 3 } )
(5): Time-Depth Separable Block (21, 80, 10) [800 -> 800 -> 800]
(6): Time-Depth Separable Block (21, 80, 10) [800 -> 800 -> 800]
(7): Conv2D (10->14, 21x1, 2,1, SAME,SAME, 1, 1) (with bias)
(8): ReLU
(9): Dropout (0.200000)
(10): LayerNorm ( axes : { 3 } )
(11): Time-Depth Separable Block (21, 80, 14) [1120 -> 1120 -> 1120]
(12): Time-Depth Separable Block (21, 80, 14) [1120 -> 1120 -> 1120]
(13): Time-Depth Separable Block (21, 80, 14) [1120 -> 1120 -> 1120]
(14): Conv2D (14->18, 21x1, 2,1, SAME,SAME, 1, 1) (with bias)
(15): ReLU
(16): Dropout (0.200000)
(17): LayerNorm ( axes : { 3 } )
(18): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(19): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(20): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(21): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(22): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(23): Time-Depth Separable Block (21, 80, 18) [1440 -> 1440 -> 1440]
(24): View (0 1440 1 0)
(25): Reorder (1,0,3,2)
(26): Linear (1440->1024) (with bias)
I0828 10:05:40.717444 985 Decode.cpp:89] [Criterion] Seq2SeqCriterion
I0828 10:05:40.717468 985 Decode.cpp:91] [Network] Number of params: 36539300
I0828 10:05:40.717474 985 Decode.cpp:97] [Network] Updating flags from config file: seq2seq_tds_librispeech/am/baseline_dev-other.binMy tree for the MODEL_DST:
seq2seq_tds_librispeech/
|-- am
| |-- baseline_dev-clean.bin
| |-- baseline_dev-other.bin
| |-- lexicon-train+dev-unigram-10000-nbest10.lexicon
| |-- tokens-train-all-unigram-10000.model
| |-- tokens-train-all-unigram-10000.tokens
| |-- tokens-train-all-unigram-10000.vocab
| `-- train.txt
`-- decoder
|-- lm_librispeech_convlm_14B.bin
|-- lm_librispeech_convlm_14B.vocab
`-- lm_librispeech_kenlm_4g.binand for the DATA_DST:
librispeech/
|-- audio
| `-- LibriSpeech
| |-- dev-clean
| |-- dev-other
| |-- test-clean
| |-- test-other
| |-- train-clean-100
| |-- train-clean-360
| `-- train-other-500
|-- lists
| |-- dev-clean.lst
| |-- dev-other.lst
| |-- test-clean.lst
| |-- test-other.lst
| |-- train-clean-100.lst
| |-- train-clean-360.lst
| `-- train-other-500.lst
`-- text
|-- dev-clean.txt
|-- dev-other.txt
|-- librispeech-lm-norm.txt
|-- librispeech-lm-norm.txt.lower.shuffle
|-- test-clean.txt
|-- test-other.txt
|-- train-clean-100.txt
|-- train-clean-360.txt
`-- train-other-500.txtand finally my configuration file (which is placed in the same level as the above 2 folders):
Decoding config for Librispeech using Seq2Seq TDS model
# Replace `[...]`, `[DATA_DST]`, `[MODEL_DST]` with appropriate paths
# for test-other (best params for dev-other)
--listdata=true
--lm=seq2seq_tds_librispeech/decoder/lm_librispeech_convlm_14B.bin
--am=seq2seq_tds_librispeech/am/baseline_dev-other.bin
--tokensdir=seq2seq_tds_librispeech/am
--test=librispeech/lists/test-other.lst
--sclite=./
--decodertype=tkn
--lmweight=0.5
--wordscore=1.0
--beamsize=80
--beamthreshold=7
--hardselection=1.5
--softselection=10.0
--attentionthreshold=30
--nthread_decoder=8
--smearing=max
--show=true
--lmtype=convlm
--lm_memory=3000
--lm_vocab=seq2seq_tds_librispeech/decoder/lm_librispeech_convlm_14B.vocab
tlikhomanenko
nimz
I downloaded the docker image for wav2letter and updated it with the latest version of this repo. I am trying to use the results from the Seq2Seq TDS model and while
recipes/models/seq2seq_tds/librispeech/prepare.pyruns just fine,build/Decoderwith therecipes/models/seq2seq_tds/librispeech/decode_convlm_other.cfghas the following error: