Arun
commit from arun
ff9a56f
raw
history blame
9.51 kB
config: conf/tuning/train_conformer-rnn_transducer.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_IndEng188h-rnnt-600bpe
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 0
dist_backend: nccl
dist_init_method: env://
dist_world_size: null
dist_rank: null
local_rank: 0
dist_master_addr: null
dist_master_port: null
dist_launcher: null
multiprocessing_distributed: false
unused_parameters: false
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 20
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
- - valid
- loss
- min
keep_nbest_models: 3
nbest_averaging_interval: 0
grad_clip: 5.0
grad_clip_type: 2.0
grad_noise: false
accum_grad: 4
no_forward_run: false
resume: true
train_dtype: float32
use_amp: false
log_interval: null
use_matplotlib: true
use_tensorboard: true
use_wandb: false
wandb_project: null
wandb_id: null
wandb_entity: null
wandb_name: null
wandb_model_log_interval: -1
detect_anomaly: false
pretrain_path: null
init_param: []
ignore_init_mismatch: false
freeze_param: []
num_iters_per_epoch: null
batch_size: 20
valid_batch_size: null
batch_bins: 64
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_en_bpe600/train/speech_shape
- exp/asr_stats_raw_en_bpe600/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_en_bpe600/valid/speech_shape
- exp/asr_stats_raw_en_bpe600/valid/text_shape.bpe
batch_type: folded
valid_batch_type: null
fold_length:
- 80000
- 150
sort_in_batch: descending
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
train_data_path_and_name_and_type:
- - dump/raw/train_English/wav.scp
- speech
- sound
- - dump/raw/train_English/text
- text
- text
valid_data_path_and_name_and_type:
- - dump/raw/dev_English/wav.scp
- speech
- sound
- - dump/raw/dev_English/text
- text
- text
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
valid_max_cache_size: null
optim: adam
optim_conf:
lr: 0.0008
weight_decay: 1.0e-06
scheduler: warmuplr
scheduler_conf:
warmup_steps: 25000
token_list:
- <blank>
- <unk>
- S
- ▁THE
- ▁A
- E
- T
- ▁TO
- I
- D
- ▁P
- ING
- ▁IN
- ▁AT
- ▁AND
- ED
- P
- ▁FOR
- A
- ▁M
- ▁OF
- Y
- O
- ER
- M
- ▁I
- U
- N
- C
- AR
- AL
- ▁C
- G
- ▁S
- ▁ON
- R
- B
- ▁F
- ▁IS
- H
- L
- ▁T
- F
- OR
- ▁TABLE
- IN
- AN
- ▁BOOK
- ▁RE
- K
- ▁THAT
- RE
- ▁BE
- ▁TWO
- IC
- IT
- LE
- EN
- V
- RA
- ▁ONE
- ▁G
- ▁D
- UR
-
- ▁IT
- LY
- ES
- 'ON'
- RI
- ▁HE
- W
- ▁YOU
- ▁WITH
- ▁B
- ▁ST
- TH
- ▁E
- ▁WE
- AT
- ▁ARE
- ▁WAS
- ▁NOT
- ▁SO
- ▁AS
- IL
- ▁WANT
- ▁W
- ENT
- ▁K
- ▁HAVE
- ▁H
- CH
- ▁THREE
- ▁DE
- RO
- AS
- ▁WILL
- ▁L
- TER
- LA
- ▁FOUR
- ATION
- ▁HAS
- LL
- ST
- ▁THIRTY
- ▁CAN
- ▁DO
- AD
- ERS
- EL
- ▁O
- AM
- ▁THIS
- UL
- IR
- ▁FROM
- TION
- ▁HIS
- ▁AN
- ION
- ▁FIVE
- ▁SE
- ▁V
- US
- ▁ME
- ▁BY
- ▁LIKE
- IS
- ▁SAID
- ▁R
- ▁EIGHT
- ▁N
- ▁CON
- ▁PEOPLE
- LI
- ▁J
- AND
- ▁BUT
- J
- ▁WHAT
- ▁TODAY
- UN
- ATE
- ▁SIX
- MENT
- ▁RA
- ▁OR
- UT
- ▁SU
- CE
- ▁WHICH
- ▁CO
- ▁EX
- NE
- OL
- Z
- ▁SEVEN
- SE
- NG
- ▁MO
- ▁NEED
- VER
- ▁ALL
- ITY
- VE
- ▁LE
- ▁GO
- ▁WHO
- GE
- IVE
- ▁MALL
- HI
- ▁AM
- AC
- IDE
- ▁WOULD
- LO
- ▁NINE
- ▁BA
- ▁MA
- UM
- ▁NEW
- ▁WHEN
- ▁TIME
- AY
- ▁DINNER
- ▁RO
- IES
- CK
- ▁ALSO
- ▁PRO
- ▁EN
- OW
- ▁VI
- OT
- ▁THERE
- ▁MY
- ▁YEAR
- ▁LA
- ▁TEN
- ▁ITS
- IM
- ▁INDIA
- ▁US
- ▁ABOUT
- ▁UP
- MA
- ▁THOUSAND
- ▁OUT
- AP
- ▁BO
- AGE
- ▁PLAY
- ▁THEY
- X
- ▁GET
- EST
- ▁THEIR
- ▁HAD
- IST
- ▁NO
- ▁SP
- ▁AFTER
- ▁YOUR
- TING
- CI
- ▁NOVEMBER
- Q
- ▁DIS
- ▁BEEN
- ▁APP
- ▁HO
- KE
- ▁UN
- OS
- ARD
- UD
- ▁OVER
- ▁PLEASE
- ACH
- ▁OKAY
- ▁PRE
- ▁MORE
- ▁SOME
- ▁SH
- NA
- ▁BR
- ANT
- ▁FIRST
- ▁PART
- TED
- OM
- ALLY
- ▁CHA
- ▁ADULTS
- RY
- URE
- IF
- ▁TWENTY
- ▁DAY
- AV
- ▁WORK
- ▁TOMORROW
- HA
- TA
- ONE
- ENCE
- ▁POINT
- ▁WERE
- ▁HA
- ▁IF
- OUS
- PE
- ▁NA
- ▁COM
- ▁HER
- ▁OTHER
- TS
- ▁NOW
- HO
- ▁ACT
- ▁HUNDRED
- ANCE
- ▁MEMBERS
- ▁HOW
- ▁SHE
- ▁QU
- ▁U
- ▁SAY
- ▁SC
- AK
- ▁KNOW
- ▁SA
- ▁AR
- ▁ANY
- ▁THAN
- ▁JO
- ▁JUST
- ▁ROAD
- MAN
- ▁TEST
- ABLE
- HE
- ▁MAN
- ▁LUNCH
- ▁TAKE
- ▁INTO
- ▁LI
- ▁WANTED
- AB
- UND
- GH
- MO
- ▁RUN
- ▁ONLY
- ▁SHA
- PER
- ITE
- ▁TEAM
- ISH
- ▁MOST
- ▁EVEN
- ▁TWELVE
- ▁RESERVE
- ▁DID
- ▁OUR
- ▁HIM
- ▁COMP
- ARY
- ▁LOOK
- ▁SEE
- ▁OFF
- ATED
- ▁CALL
- ▁SECOND
- ▁WHILE
- WAR
- ▁MAKE
- ▁FILM
- ▁WIN
- ▁WORLD
- ▁RIGHT
- TEN
- ▁HMM
- ▁PLACE
- ▁STATE
- ▁INDIAN
- ▁VERY
- ▁REPORT
- ▁COULD
- ▁LAST
- ▁THINK
- ▁START
- ▁ADD
- ▁NATION
- ▁WELL
- ▁HIGH
- ▁WHERE
- ▁MADE
- ▁MOVIE
- ▁BACK
- ▁COME
- ASH
- ▁EVERY
- ▁KA
- ▁BEST
- ▁BECAUSE
- ▁ENGLAND
- ICK
- ▁SHOW
- ▁WAY
- COM
- ▁GIVE
- ▁KIND
- ▁CHILDREN
- ▁BEFORE
- ▁PAKISTAN
- ▁LEAD
- ▁WATCH
- ▁THROUGH
- ▁MUCH
- ▁MANY
- ▁GOOD
- ▁SUNDAY
- ▁FIFTY
- PORT
- ▁NAGAR
- ▁THIRD
- ▁GREAT
- ▁UNDER
- ▁ADULT
- ▁HELP
- ▁MONDAY
- LIGHT
- ▁USERS
- ▁SHOULD
- ▁BAN
- ▁BALL
- ▁PLAN
- ▁VIDEO
- ▁MONTH
- ▁Z
- LINE
- ▁NEXT
- ▁DIRECT
- ▁BIG
- ▁YES
- ▁CAME
- ▁MUMBAI
- ▁SEATS
- ▁MATCH
- ▁LOT
- ▁GAME
- ▁PERSON
- ▁AGAINST
- ▁WHY
- ▁IMP
- ▁OLD
- ▁RESERVATION
- ▁COMPANY
- ▁DURING
- ▁FEATURE
- ▁COMMUN
- ▁BETWEEN
- ▁WICKET
- ▁SCORE
- ▁LIFE
- ▁DOWN
- ▁SATURDAY
- OOD
- ▁TALK
- ▁HOME
- ▁OPEN
- ▁YEAH
- ▁CHANGE
- ▁NINETEEN
- ▁SIXTY
- ▁ELECT
- ▁NINETY
- ▁LAUNCH
- ▁TRANS
- ▁FINAL
- ▁CRICKET
- ▁GOVERNMENT
- ▁FOUND
- ▁NUMBER
- ▁HOWEVER
- ▁ZERO
- ▁SINCE
- ▁DIFFERENT
- ▁SERIES
- ▁MUSIC
- ▁DELHI
- ▁SUPPORT
- ▁SEVENTY
- ▁AUSTRALIA
- ▁SIXTEEN
- ▁CHECK
- ▁MAKING
- ▁MINUTE
- ▁THINGS
- ▁CHENNAI
- ▁AROUND
- ▁FRIDAY
- ▁LONG
- ▁THURSDAY
- ▁GOOGLE
- ▁MINISTER
- ▁BOWL
- ▁SAME
- ▁KHAN
- ▁SPACE
- ▁FACEBOOK
- ▁BARB
- ▁POST
- ▁FOLLOW
- ▁WEEK
- ▁RESEARCH
- AUGHT
- ▁SPEAK
- ▁POLICE
- ▁INNINGS
- ▁BECOME
- ▁KALYAN
- ▁FEEL
- ▁RELEASE
- ▁THOSE
- ▁MILLION
- ▁ISSUE
- ▁PHONE
- ▁FIND
- ▁RAHUL
- ▁PROBLEM
- ▁STILL
- ▁COURT
- ▁HAPPEN
- ▁KEEP
- ▁ELSE
- ▁ANOTHER
- ▁DEVELOP
- ▁CREAT
- ▁WEDNESDAY
- ▁TELL
- ▁CLOSE
- ▁SOUTH
- ▁TURN
- ▁THOUGH
- ▁TUESDAY
- ▁KOHLI
- ▁SOMETHING
- ▁CHILD
- ▁LOVE
- ▁ACCORDING
- ▁DEVICE
- ▁IMPORTANT
- ▁GROUP
- ▁CONTINUE
- ▁FAMILY
- ▁DRIVE
- ▁COUNTRY
- ▁CAPTAIN
- ▁COUNTRIES
- ▁WATER
- ▁OWN
- ▁PROJECT
- ▁DECEMBER
- ▁WITHOUT
- ▁CENTURY
- ▁FIFTEEN
- ▁OFFICE
- ▁ENGLISH
- ▁PROGRAM
- ▁FRIEND
- ▁RUPEES
- ▁CONSIDER
- ▁BENGALURU
- ▁PERFORM
- ▁ACCOUNT
- ▁PRIME
- ▁MEDIA
- ▁UNDERSTAND
- ▁EXPECT
- ▁RECORD
- ▁EXPERIENCE
- ▁OFFICIAL
- ▁DECISION
- ▁ELEVEN
- ▁TONIGHT
- ▁ATTACK
- ▁BREAK
- ▁AGAIN
- ▁INCLUDING
- ▁DESIGN
- ▁ANNOUNCE
- ▁COMPLETE
- ▁SERVICE
- ▁SEVENTEEN
- ▁ALREADY
- ▁TRAVEL
- ▁PRODUCT
- ▁PROCESS
- ▁PRESIDENT
- ▁KOLKATA
- ▁SYSTEM
- ▁ALWAYS
- ▁SMARTPHONE
- ▁SUCCESS
- ▁MAJOR
- ▁PHOTO
- ▁IMPROVE
- ▁AMERICA
- ▁SQUARE
- ▁LEAGUE
- ▁INTERNATIONAL
- ▁BUILD
- ▁ACTUALLY
- ▁SIGN
- ▁CHIEF
- ▁TWITTER
- '1'
- '0'
- '3'
- '4'
- '9'
- '2'
- i
- o
- <
- s
- ''''
- /
- '5'
- '7'
- '8'
- '>'
- e
- '6'
- <sos/eos>
init: null
input_size: null
ctc_conf:
dropout_rate: 0.0
ctc_type: builtin
reduce: true
ignore_nan_grad: true
joint_net_conf:
joint_space_size: 640
use_preprocessor: true
token_type: bpe
bpemodel: data/en_token_list/bpe_unigram600/bpe.model
non_linguistic_symbols: null
cleaner: null
g2p: null
speech_volume_normalize: null
rir_scp: null
rir_apply_prob: 1.0
noise_scp: null
noise_apply_prob: 1.0
noise_db_range: '13_15'
frontend: default
frontend_conf:
n_fft: 512
hop_length: 160
fs: 16k
specaug: specaug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2
normalize: global_mvn
normalize_conf:
stats_file: exp/asr_stats_raw_en_bpe600/train/feats_stats.npz
model: espnet
model_conf:
ctc_weight: 0.3
report_cer: false
report_wer: false
preencoder: null
preencoder_conf: {}
encoder: conformer
encoder_conf:
output_size: 512
attention_heads: 8
linear_units: 2048
num_blocks: 12
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: conv2d
normalize_before: true
macaron_style: true
rel_pos_type: latest
pos_enc_layer_type: rel_pos
selfattention_layer_type: rel_selfattn
activation_type: swish
use_cnn_module: true
cnn_module_kernel: 15
postencoder: null
postencoder_conf: {}
decoder: transducer
decoder_conf:
rnn_type: lstm
num_layers: 1
hidden_size: 512
dropout: 0.1
dropout_embed: 0.2
required:
- output_dir
- token_list
version: '202205'
distributed: false