Hugging Face's logo

popcornell
/
chime7_task1_asr1_baseline

Automatic Speech Recognition
ESPnet
English
audio
Model card Files Community
Edit model card

ESPnet2 ASR model

popcornell/chime7_task1_asr1_baseline

This model was trained by popcornell using chime7_task1 recipe in espnet.

Demo: How to use in ESPnet2

Follow the ESPnet installation instructions if you haven't done that already.

cd espnet
git checkout 89ebca463c544dfaa19e5f76ad5f615f473f6957
pip install -e .
cd egs2/chime7_task1/asr1
./run.sh --skip_data_prep false --skip_train true --download_model popcornell/chime7_task1_asr1_baseline

RESULTS

Environments

  • date: Mon Mar 13 13:43:21 UTC 2023
  • python version: 3.9.2 (default, Mar 3 2021, 20:02:32) [GCC 7.3.0]
  • espnet version: espnet 202301
  • pytorch version: pytorch 1.13.1
  • Git hash: 89ebca463c544dfaa19e5f76ad5f615f473f6957
    • Commit date: Tue Mar 7 04:02:43 2023 +0000

exp/kaldi/mixer6/gss

WER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 14804 148981 84.4 10.4 5.2 6.2 21.8 60.6

CER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 14804 731649 91.0 3.6 5.4 6.5 15.5 60.6

TER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 14804 251635 87.0 7.4 5.6 6.9 19.9 60.6

exp/kaldi/chime6/gss_inf

WER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss_inf 6644 58881 73.0 18.2 8.8 6.7 33.6 71.2

CER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss_inf 6644 281489 83.2 7.0 9.8 7.3 24.1 71.2

TER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss_inf 6644 98596 75.0 14.2 10.9 7.1 32.2 71.2

exp/kaldi/dipco/gss

WER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 3673 29966 74.0 17.8 8.2 8.5 34.5 72.6

CER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 3673 146438 84.5 6.3 9.3 8.9 24.4 72.6

TER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
dev/gss 3673 51347 77.1 13.5 9.4 9.2 32.1 72.6

ASR config

expand 
config: conf/tuning/train_asr_transformer_wavlm_lr1e-4_specaugm_accum1_preenc128_warmup20k.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_train_asr_transformer_wavlm_lr1e-4_specaugm_accum1_preenc128_warmup20k_raw_en_bpe500_batch_size640_scheduler_confwarmup_steps8000_max_epoch8_optim_conflr0.000500000000_sp
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 5
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 38257
dist_launcher: null
multiprocessing_distributed: true
unused_parameters: true
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 8
patience: 4
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 5
nbest_averaging_interval: 0
grad_clip: 5
grad_clip_type: 2.0
grad_noise: false
accum_grad: 1
no_forward_run: false
resume: true
train_dtype: float32
use_amp: false
log_interval: null
use_matplotlib: true
use_tensorboard: true
create_graph_in_tensorboard: false
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:
- frontend.upstream
num_iters_per_epoch: null
batch_size: 640
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_en_bpe500_sp/train/speech_shape
- exp/asr_stats_raw_en_bpe500_sp/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_en_bpe500_sp/valid/speech_shape
- exp/asr_stats_raw_en_bpe500_sp/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
chunk_excluded_key_prefixes: []
train_data_path_and_name_and_type:
-   - dump/raw/kaldi/train_all_mdm_ihm_rvb_gss_sp/wav.scp
    - speech
    - sound
-   - dump/raw/kaldi/train_all_mdm_ihm_rvb_gss_sp/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/kaldi/chime6/dev/gss/wav.scp
    - speech
    - sound
-   - dump/raw/kaldi/chime6/dev/gss/text
    - text
    - text
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
valid_max_cache_size: null
exclude_weight_decay: false
exclude_weight_decay_conf: {}
optim: adam
optim_conf:
    lr: 0.0005
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 8000
token_list:
- <blank>
- <unk>
- '[inaudible]'
- '[laughs]'
- '[noise]'
- s
- ''''
- ▁i
- t
- ▁it
- ▁a
- e
- ▁you
- ▁the
- ▁like
- ▁yeah
- a
- d
- ▁and
- m
- ▁that
- ▁to
- n
- i
- y
- ing
- o
- u
- ▁so
- p
- ▁of
- ▁in
- re
- ▁was
- c
- r
- ▁just
- er
- ▁know
- ▁oh
- ed
- ▁but
- ▁ummm
- ▁we
- l
- ▁no
- ▁they
- ▁have
- ▁do
- g
- ▁he
- k
- ll
- ▁uhhh
- ▁don
- ▁for
- h
- ▁what
- ▁be
- ar
- ▁is
- ▁there
- '-'
- ▁s
- ▁this
- in
- b
- ▁
- en
- ▁on
- ▁p
- ▁can
- al
- ▁not
- w
- ▁my
- ▁one
- ic
- f
- ▁or
- ▁really
- ▁go
- ▁right
- ▁me
- an
- ▁w
- or
- le
- ▁f
- ▁think
- ▁okay
- ▁all
- ▁then
- ▁with
- ▁are
- ▁get
- it
- ▁t
- ▁st
- ve
- ▁hmmm
- ▁g
- ▁if
- ce
- 'on'
- ▁she
- ▁good
- ▁e
- es
- ▁well
- v
- ▁re
- th
- ter
- ch
- ▁out
- ▁up
- ly
- ▁b
- ▁ma
- il
- ▁would
- ▁at
- ▁want
- ▁mean
- ▁ch
- ▁your
- ▁people
- ur
- ▁how
- ▁k
- ▁co
- ▁about
- ▁tr
- ▁ba
- ▁kind
- ▁when
- ▁mi
- ▁because
- ro
- ▁had
- ▁ho
- ▁gonna
- ▁time
- ▁more
- ▁got
- ▁some
- ▁two
- ▁did
- ▁see
- ▁now
- ▁pa
- ra
- ▁de
- ▁lot
- ▁actually
- ▁o
- ▁too
- ate
- ▁here
- ▁cuz
- ▁sp
- ▁where
- ▁going
- ▁j
- ▁from
- ▁bo
- ▁them
- ▁bu
- ▁put
- ▁thing
- ng
- ▁were
- ▁n
- ▁sh
- ▁work
- el
- ▁something
- ▁se
- ▁say
- ke
- ow
- ▁ca
- ▁fa
- ▁need
- sh
- ▁di
- ▁po
- ▁make
- la
- ▁br
- ▁v
- ▁an
- ▁who
- ion
- ▁y
- ▁look
- ▁didn
- ▁could
- ▁little
- ver
- ▁c
- ▁mo
- ▁much
- ▁very
- ir
- ▁sa
- ▁play
- ▁pretty
- ▁been
- ▁d
- ▁other
- ▁year
- and
- ▁mm
- ▁stuff
- ▁dr
- ▁why
- ▁con
- ▁su
- ▁back
- ▁ex
- ting
- ▁take
- ▁li
- ▁even
- ▁should
- ▁her
- ally
- lo
- ation
- ▁way
- ▁guess
- ▁has
- z
- ▁three
- ry
- ▁ha
- ies
- is
- x
- ▁ro
- ▁yes
- ▁th
- ▁use
- ▁down
- ous
- ▁over
- ▁probably
- ▁guys
- ▁maybe
- ▁still
- ▁cr
- ▁which
- ▁nice
- und
- ▁sure
- ▁l
- ▁off
- ▁la
- ▁cu
- est
- ▁any
- ▁fi
- ▁these
- ▁ra
- ▁went
- ▁things
- ment
- ▁doing
- ▁day
- ▁un
- ▁lo
- ▁da
- ▁only
- igh
- ▁come
- ▁big
- ▁those
- ▁wanna
- ▁bit
- ▁never
- ▁us
- ol
- ▁though
- ▁first
- ive
- ▁their
- ▁let
- ▁start
- ▁his
- ▁four
- ▁le
- ▁eat
- ist
- ▁school
- us
- ▁into
- ▁yep
- uck
- ▁than
- ▁him
- ▁hi
- ▁also
- ▁five
- side
- ▁new
- ▁comp
- ▁cool
- ▁talk
- ▁said
- ▁pro
- ▁r
- ▁always
- ▁ri
- ▁cl
- ▁long
- able
- ▁sc
- ▁gra
- ▁by
- ▁friend
- age
- ▁different
- ▁live
- ▁doesn
- ▁place
- ▁sorry
- ▁will
- ▁feel
- ▁does
- ▁part
- ▁wait
- ▁six
- ▁watch
- ▁anything
- ▁man
- ▁our
- ▁car
- ▁huh
- ▁whatever
- ▁last
- ▁give
- ▁ten
- ▁before
- ▁thought
- ▁after
- ▁game
- ▁card
- ▁fl
- ▁every
- cause
- ▁same
- ▁around
- ▁cook
- ▁week
- ▁hu
- ▁everything
- ▁fine
- ▁many
- ▁qu
- ▁read
- ▁tea
- ough
- ance
- ▁turn
- ▁wow
- ▁fun
- ▁hard
- ▁great
- ▁love
- ▁remember
- ▁twenty
- ▁whole
- ▁happen
- ▁seven
- ▁keep
- ▁food
- ▁most
- j
- ▁might
- ▁thank
- ▁move
- ▁job
- ▁eight
- ▁mu
- ▁sort
- ▁better
- port
- ▁another
- ful
- ▁point
- ▁show
- ▁again
- ▁high
- ize
- ▁house
- ▁home
- ▁person
- ▁old
- ▁end
- ▁through
- ▁pick
- ▁else
- ▁guy
- ▁app
- ▁find
- ▁nine
- ▁hand
- ▁kid
- ▁interesting
- ▁city
- ▁called
- ▁tell
- ▁half
- ▁name
- ▁definitely
- ▁made
- ▁exactly
- ▁came
- ▁wood
- ▁funny
- ▁basically
- ▁count
- ▁usually
- ▁help
- ▁someone
- ▁already
- ▁dunno
- ▁enough
- ction
- ▁own
- ▁weird
- ▁next
- ▁hundred
- ▁small
- ▁money
- ▁couple
- ▁while
- ▁close
- ▁movie
- ▁sometimes
- ▁everyone
- ▁away
- ▁true
- ▁super
- ▁cheese
- ▁class
- ▁night
- ▁life
- ▁leave
- ▁plan
- ▁water
- ▁left
- ▁thirty
- ▁family
- ▁phone
- ▁build
- ▁room
- ▁month
- ▁open
- ▁idea
- ▁second
- ▁dude
- ▁music
- ▁each
- ▁learn
- ▁girl
- ▁together
- ▁under
- ▁run
- ▁chicken
- ▁having
- ▁either
- ▁almost
- ▁crazy
- ▁book
- ▁sauce
- ▁supposed
- ▁course
- ▁speak
- ▁awesome
- ▁anyway
- ▁throw
- ▁finish
- ▁world
- ▁reason
- ▁check
- ▁least
- '&'
- ä
- '#'
- ñ
- â
- é
- ü
- î
- ']'
- q
- <sos/eos>
init: xavier_uniform
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: null
    zero_infinity: true
joint_net_conf: null
use_preprocessor: true
token_type: bpe
bpemodel: data/en_token_list/bpe_unigram500/bpe.model
non_linguistic_symbols: data/nlsyms.txt
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'
short_noise_thres: 0.5
aux_ctc_tasks: []
frontend: s3prl
frontend_conf:
    frontend_conf:
        upstream: wavlm_large
    download_dir: ./hub
    multilayer_feature: true
    fs: 16k
specaug: specaug
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: false
    freq_mask_width_range:
    - 0
    - 150
    num_freq_mask: 4
    apply_time_mask: true
    time_mask_width_ratio_range:
    - 0.0
    - 0.15
    num_time_mask: 3
normalize: utterance_mvn
normalize_conf: {}
model: espnet
model_conf:
    ctc_weight: 0.3
    lsm_weight: 0.1
    length_normalized_loss: false
    extract_feats_in_collect_stats: false
preencoder: linear
preencoder_conf:
    input_size: 1024
    output_size: 128
    dropout: 0.2
encoder: transformer
encoder_conf:
    output_size: 256
    attention_heads: 4
    linear_units: 2048
    num_blocks: 12
    dropout_rate: 0.1
    attention_dropout_rate: 0.0
    input_layer: conv2d2
    normalize_before: true
postencoder: null
postencoder_conf: {}
decoder: transformer
decoder_conf:
    input_layer: embed
    attention_heads: 4
    linear_units: 2048
    num_blocks: 6
    dropout_rate: 0.1
    positional_dropout_rate: 0.0
    self_attention_dropout_rate: 0.0
    src_attention_dropout_rate: 0.0
preprocessor: default
preprocessor_conf: {}
required:
- output_dir
- token_list
version: '202301'
distributed: true

Citing ESPnet 

@inproceedings{watanabe2018espnet,
  author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
  title={{ESPnet}: End-to-End Speech Processing Toolkit},
  year={2018},
  booktitle={Proceedings of Interspeech},
  pages={2207--2211},
  doi={10.21437/Interspeech.2018-1456},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
}

or arXiv:

@misc{watanabe2018espnet,
  title={ESPnet: End-to-End Speech Processing Toolkit}, 
  author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
  year={2018},
  eprint={1804.00015},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}
Downloads last month
0