Hugging Face's logo

espnet
/
universa-wavlm_base_urgent24_multi-metric_fullref

ESPnet
multilingual
audio
universa
Model card Files Community
universa-wavlm_base_urgent24_multi-metric_fullref / README.md
ftshijt's picture
ftshijt
Update model
a9a0fbe
preview code
|
raw
history blame contribute delete
10.8 kB
---
tags:
- espnet
- audio
- universa
language: multilingual
datasets:
- urgent24
license: cc-by-4.0
---
## ESPnet2 universa model
### `espnet/universa-wavlm_base_urgent24_multi-metric_fullref`
This model was trained by ftshijt using urgent24 recipe in [espnet](https://github.com/espnet/espnet/).
### Demo: How to use in ESPnet2
Follow the [ESPnet installation instructions](https://espnet.github.io/espnet/installation.html)
if you haven't done that already.
```bash
cd espnet
git checkout 5dbbd4cec6d4ccdd840484207951770027a8d4b8
pip install -e .
cd egs2/urgent24/uni_versa1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/universa-wavlm_base_urgent24_multi-metric_fullref
```
## universa config
<details><summary>expand</summary>
```
config: conf/train_universa_wavlm_freeze.yaml
print_config: false
log_level: INFO
drop_last_iter: false
dry_run: false
iterator_type: sequence
valid_iterator_type: null
output_dir: update_exp/universa_train_universa_wavlm_freeze_raw_fs16000
ngpu: 1
seed: 777
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
use_deepspeed: false
deepspeed_config: null
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: false
use_tf32: false
collect_stats: false
write_collected_feats: false
max_epoch: 100
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
- - train
- loss
- min
- - valid
- loss
- min
- - train
- acc
- max
- - valid
- acc
- max
keep_nbest_models: 5
nbest_averaging_interval: 0
grad_clip: -1
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: 50
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
use_adapter: false
adapter: lora
save_strategy: all
adapter_conf: {}
pretrain_path: null
init_param: []
ignore_init_mismatch: false
freeze_param:
- frontend.upstream
num_iters_per_epoch: null
batch_size: 16
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
category_sample_size: 10
train_shape_file:
- update_exp/universa_stats_raw/train/audio_shape
- update_exp/universa_stats_raw/train/ref_audio_shape
- update_exp/universa_stats_raw/train/ref_text_shape
valid_shape_file:
- update_exp/universa_stats_raw/valid/audio_shape
- update_exp/universa_stats_raw/valid/ref_audio_shape
- update_exp/universa_stats_raw/valid/ref_text_shape
batch_type: sorted
valid_batch_type: null
fold_length:
- 256000
sort_in_batch: descending
shuffle_within_batch: false
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
chunk_excluded_key_prefixes: []
chunk_default_fs: null
chunk_max_abs_length: null
chunk_discard_short_samples: true
train_data_path_and_name_and_type:
- - dump_ark/raw/train_update/wav.scp
- audio
- kaldi_ark
- - dump_ark/raw/train_update/metric.scp
- metrics
- metric
- - dump_ark/raw/train_update/ref_wav.scp
- ref_audio
- kaldi_ark
- - dump_ark/raw/train_update/text
- ref_text
- text
valid_data_path_and_name_and_type:
- - dump_ark/raw/dev_update/wav.scp
- audio
- kaldi_ark
- - dump_ark/raw/dev_update/metric.scp
- metrics
- metric
- - dump_ark/raw/dev_update/ref_wav.scp
- ref_audio
- kaldi_ark
- - dump_ark/raw/dev_update/text
- ref_text
- text
multi_task_dataset: false
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
allow_multi_rates: false
valid_max_cache_size: null
exclude_weight_decay: false
exclude_weight_decay_conf: {}
optim: adamw
optim_conf:
lr: 0.001
scheduler: warmuplr
scheduler_conf:
warmup_steps: 25000
metric2id: dump_ark/raw/train_update/metric2id
metric2type: null
metric_pad_value: -100
token_list:
- <blank>
- <unk>
- s
- ▁
- t
- e
- ▁the
- i
- a
- o
- ▁a
- r
- ▁to
- d
- ▁and
- ''''
- m
- n
- ing
- u
- y
- p
- c
- ▁of
- l
- ed
- ▁I
- ▁in
- er
- re
- ▁it
- ▁you
- ar
- ▁f
- ▁is
- ▁that
- ','
- .
- in
- al
- g
- 'on'
- ▁b
- b
- or
- ▁c
- ▁s
- f
- h
- ▁we
- an
- en
- ▁for
- le
- ▁p
- ly
- es
- w
- ▁re
- ▁on
- ▁m
- ▁be
- ic
- ll
- th
- ▁he
- k
- ur
- ve
- ▁with
- ▁so
- ▁from
- ▁was
- v
- ch
- st
- ▁w
- ▁i
- ▁this
- ▁de
- ▁like
- ▁do
- ce
- at
- il
- ck
- ▁A
- ▁have
- ▁not
- ad
- ▁st
- ow
- ro
- ne
- ▁me
- ▁my
- ▁but
- ation
- ▁at
- ▁or
- '-'
- ter
- ent
- ▁B
- ▁n
- ▁know
- ▁t
- out
- ▁are
- nd
- ▁one
- ▁li
- ▁g
- ▁The
- ol
- ion
- te
- ▁go
- ut
- ▁as
- ▁just
- as
- ▁sh
- ▁they
- is
- ▁C
- et
- ▁h
- ▁an
- ▁there
- ▁up
- ▁S
- ▁M
- ▁she
- ▁by
- ▁su
- om
- ▁can
- us
- ▁your
- ng
- ▁con
- el
- ▁us
- ment
- z
- ▁see
- ▁ab
- ▁what
- ▁out
- ▁her
- me
- ate
- ▁all
- ▁th
- ▁if
- ▁right
- ▁his
- ▁ma
- ▁lo
- ▁which
- ide
- ▁P
- ▁more
- ▁then
- ul
- ast
- x
- ight
- ill
- ▁So
- ▁sp
- ▁going
- ▁some
- ure
- ▁their
- ig
- ▁no
- ▁ro
- ▁think
- ▁who
- ▁pro
- ver
- ive
- est
- ▁co
- ▁di
- '0'
- ist
- ▁k
- age
- ▁d
- ▁time
- ▁L
- ies
- ▁will
- ▁man
- ▁when
- ▁D
- les
- ▁F
- ▁want
- ff
- ity
- ▁un
- '?'
- ▁start
- ▁G
- ▁uh
- ▁get
- ok
- ▁take
- ▁po
- li
- ▁ho
- ▁way
- ▁don
- ▁yeah
- ▁really
- ▁say
- ▁look
- ▁good
- ▁ra
- ▁pr
- ▁had
- ttle
- ▁comp
- ort
- ish
- ▁ex
- ally
- ▁sa
- ▁how
- end
- ant
- ▁O
- ▁um
- way
- ance
- ▁other
- ▁two
- ine
- ever
- able
- ▁com
- other
- ▁first
- ▁back
- ▁al
- ers
- ions
- ▁now
- ▁off
- ning
- ▁down
- ▁has
- ▁than
- ▁car
- ▁Th
- very
- ice
- ▁dr
- ▁been
- ▁him
- ▁here
- ated
- '5'
- ▁hand
- ▁day
- ▁hear
- each
- ▁would
- ▁over
- ▁oh
- ▁cha
- ood
- ▁did
- ugh
- ▁per
- ▁let
- ▁str
- ▁tra
- ▁got
- ext
- '1'
- ▁We
- ▁Shields
- ▁come
- ▁should
- ▁could
- light
- '2'
- ▁people
- ▁again
- ▁year
- ▁app
- ▁into
- ▁any
- ▁N
- ▁mean
- ▁o
- ▁mus
- ▁lot
- ▁said
- ▁long
- ▁these
- ▁lea
- sh
- ▁vi
- ▁part
- ▁every
- ▁our
- ▁You
- ious
- ▁fight
- ▁Ch
- ark
- ▁may
- ▁Hammer
- ▁because
- ▁most
- ▁came
- ▁four
- ful
- ▁No
- ize
- ▁where
- ▁okay
- ▁much
- ▁ask
- ▁through
- ▁before
- ▁work
- ▁even
- ▁three
- mber
- ▁win
- ▁flight
- ake
- K
- ▁place
- ▁play
- ▁though
- ▁pound
- ▁bit
- land
- ▁va
- ▁talk
- ▁kind
- ▁Line
- ▁make
- hap
- ▁big
- ▁leav
- ▁something
- ▁game
- ▁under
- ▁feel
- self
- ▁give
- ▁includ
- U
- ▁twenty
- ▁guard
- ▁left
- ▁round
- ▁great
- body
- ▁gra
- ress
- lso
- '3'
- ▁everything
- ▁those
- ▁after
- ▁tell
- ▁need
- ▁yes
- qua
- ham
- ▁minutes
- ▁question
- ▁around
- ▁punch
- ▁course
- ▁gonna
- ▁person
- ▁move
- ▁plan
- ▁ear
- ept
- ▁Airport
- ▁Okay
- ▁found
- ▁seven
- ▁help
- que
- ▁qui
- ▁keep
- ▁guys
- ▁house
- ▁run
- ▁turn
- ▁better
- ▁stop
- ward
- ddle
- ▁second
- ground
- ▁world
- ▁high
- ▁point
- ▁hold
- ▁call
- '6'
- ▁actually
- ▁probably
- ▁heaven
- ▁speci
- ▁everyone
- ▁why
- ▁presen
- ▁thir
- lright
- ▁eye
- eath
- ▁Tak
- '!'
- '"'
- '4'
- ▁hundred
- ▁answer
- ▁small
- ▁wait
- ▁nothing
- q
- '8'
- V
- ▁countr
- ▁problem
- ▁continu
- ▁close
- ▁priva
- ▁20
- ▁pleas
- ▁walk
- ▁open
- ▁lay
- ▁Station
- ▁moment
- ▁Yeah
- ▁public
- possibl
- ▁happen
- together
- ▁while
- asically
- ▁money
- ▁wrong
- B
- ▁puzzle
- '7'
- ▁journ
- ▁rainbow
- ▁thousand
- I
- '9'
- S
- P
- '%'
- A
- D
- L
- F
- ’
- O
- G
- N
- á
- C
- $
- Z
- Y
- R
- E
- J
- W
- M
- H
- j
- –
- ;
- Q
- X
- ']'
- −
- '&'
- T
- '['
- <sos/eos>
init: xavier_uniform
model_conf: {}
use_ref_audio: true
use_ref_text: true
use_preprocessor: true
token_type: bpe
bpemodel: data/token_list/bpe_unigram500/bpe.model
non_linguistic_symbols: null
cleaner: null
g2p: null
frontend: s3prl
frontend_conf:
frontend_conf:
upstream: wavlm_large
download_dir: ./hub
multilayer_feature: true
universa: base
universa_conf:
embedding_dim: 256
audio_encoder_type: transformer
audio_encoder_params:
num_blocks: 4
attention_heads: 4
linear_units: 1024
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: conv2d
normalize_before: true
concat_after: false
positionwise_layer_type: linear
positionwise_conv_kernel_size: 1
layer_drop_rate: 0.1
qk_norm: false
use_flash_attn: false
text_encoder_type: transformer
text_encoder_params:
num_blocks: 4
attention_heads: 4
linear_units: 1024
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: linear
normalize_before: true
concat_after: false
positionwise_layer_type: linear
positionwise_conv_kernel_size: 1
layer_drop_rate: 0.1
qk_norm: false
use_flash_attn: false
cross_attention_type: multihead
cross_attention_params:
n_head: 4
dropout_rate: 0.1
pooling_type: mean
projector_type: linear
multi_branch: true
required:
- output_dir
- metric2id
version: '202412'
distributed: false
```
</details>
### Citing ESPnet
```BibTex
@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:
```bibtex
@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}
}
```