stop_hubert_slu_raw_en_bpe500

Usage

import librosa
from espnet2.bin.asr_inference import Speech2Text

speech2text = Speech2Text.from_pretrained(model_tag="espnet/stop_hubert_slu_raw_en_bpe500")
# librosa resamples and mixes to one channel, so any file works; 16000 is
# what nearly every espnet recogniser is trained on - check this model's
# config if its audio is not 16 kHz
speech, rate = librosa.load("audio.wav", sr=16000, mono=True)
text, *_ = speech2text(speech)[0]
print(text)

ESPnet2 ASR model

espnet/stop_hubert_slu_raw_en_bpe500

This model was trained by Siddhant using stop recipe in espnet.

Demo: How to use in ESPnet2

cd espnet
git checkout 11890fdd9dd872edc50ce8eb7660d746c6ee160e
pip install -e .
cd egs2/stop/asr3
./run.sh --skip_data_prep false --skip_train true --download_model espnet/stop_hubert_slu_raw_en_bpe500

<!-- Generated by scripts/utils/show_asr_result.sh -->

RESULTS

Environments

• date: Sun Dec 25 13:33:10 EST 2022
• python version: 3.9.5 (default, Jun 4 2021, 12:28:51) [GCC 7.5.0]
• espnet version: espnet 202205
• pytorch version: pytorch 1.13.0+cu116
• Git hash: 11890fdd9dd872edc50ce8eb7660d746c6ee160e
  • Commit date: Sat Jun 18 17:05:39 2022 -0400

asr_train_asr2_hubert_lr0.002_raw_en_bpe500

WER

CER

TER

ASR config

<details><summary>expand</summary>

config: conf/train_asr2_hubert_lr0.002.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_train_asr2_hubert_lr0.002_raw_en_bpe500
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 4
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 57197
dist_launcher: null
multiprocessing_distributed: true
unused_parameters: false
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 50
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 10
nbest_averaging_interval: 0
grad_clip: 5.0
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
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: 128
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_en_bpe500/train/speech_shape
- exp/asr_stats_raw_en_bpe500/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_en_bpe500/valid/speech_shape
- exp/asr_stats_raw_en_bpe500/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/wav.scp
    - speech
    - sound
-   - dump/raw/train/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/valid/wav.scp
    - speech
    - sound
-   - dump/raw/valid/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: 0.0004
    weight_decay: 1.0e-06
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 25000
token_list:
- <blank>
- <unk>
- ▁[
- ':'
- ▁]
- _
- SL
- IN
- GET
- S
- TIME
- DATE
- ▁THE
- ▁TO
- ▁FOR
- ▁
- E
- LOCATION
- A
- WEATHER
- O
- ▁ME
- MUSIC
- ▁MY
- CREATE
- ALARM
- Y
- D
- ▁I
- T
- ▁AT
- I
- ▁A
- TIMER
- ▁IS
- U
- ▁IN
- ▁ON
- EVENT
- M
- ▁TIMER
- TODO
- REMINDER
- R
- ▁PM
- P
- ING
- ▁WHAT
- ▁THIS
- ▁TODAY
- ▁AM
- N
- ▁ALARM
- ▁SET
- NT
- METHOD
- ▁TOMORROW
- ER
- TYPE
- B
- ATTRIBUTE
- DESTINATION
- ▁MINUTES
- REMINDED
- PERSON
- L
- ▁HOW
- NAME
- K
- ▁FIVE
- ▁BE
- ▁'
- G
- ▁NEXT
- 'ON'
- ▁IT
- MESSAGE
- H
- ▁WILL
- ▁S
- ▁WEEK
- ST
- C
- INFO
- EN
- CATEGORY
- TRAFFIC
- ▁F
- LE
- ▁AND
- AR
- SEND
- RE
- ▁P
- ▁D
- ▁FROM
- RECIPIE
- PLAY
- ▁DO
- ▁TRAFFIC
- AN
- ▁AN
- AL
- ▁SIX
- ▁SONG
- ▁ALL
- ▁UP
- CONTENT
- ▁REMINDER
- ▁WEEKEND
- ▁REMIND
- ▁OF
- ▁T
- RA
- ▁WEATHER
- ▁SEVEN
- ▁PLEASE
- ▁RE
- ▁TONIGHT
- EXACT
- ▁EIGHT
- ▁W
- W
- ▁TEN
- F
- SOURCE
- ▁TIME
- ESTIMATED
- RECURRING
- TH
- DELETE
- VE
- ▁NEW
- LL
- ▁EVERY
- ▁PLAY
- ES
- ▁THIRTY
- ▁GET
- ▁RAIN
- CK
- ▁TWO
- ▁C
- ▁CO
- ▁ARE
- ▁MESSAGE
- RI
- ▁G
- ▁MORNING
- CONTACT
- ▁CAN
- ▁NOW
- ▁THREE
- ▁THERE
- ET
- ▁MUSIC
- TER
- ▁TAKE
- IC
- CH
- ▁J
- V
- ED
- ▁FOUR
- DURATION
- LY
- ▁E
- ▁FRIDAY
- UR
- ▁YOU
- ▁ANY
- ▁NINE
- ▁GO
- UNSUPPORTED
- OR
- ▁SHOW
- ▁O
- ▁BA
- ▁PA
- ▁LONG
- AT
- ▁ONE
- ND
- ▁MA
- ▁ST
- ▁GOING
- ▁LIKE
- ▁ALARMS
- ▁BY
- ▁THAT
- ▁TWENTY
- ▁DAY
- ▁CH
- ▁MONTH
- ▁K
- ▁SH
- UPDATE
- ▁MONDAY
- CE
- IT
- IL
- AMOUNT
- ▁SATURDAY
- ▁BR
- ▁NEED
- ▁WORK
- ID
- ▁DRIVE
- LA
- ▁MO
- ▁HAVE
- ▁TUESDAY
- ▁TELL
- IR
- HA
- ''''
- ▁IF
- HOME
- ▁HE
- ▁LO
- ▁LA
- ▁WHEN
- LO
- ▁TH
- ▁REMINDERS
- IE
- DISTANCE
- ▁WE
- ▁SA
- ▁HOUR
- OULD
- NE
- DEPARTURE
- ▁HI
- ▁LI
- ARTIST
- Z
- TRAVEL
- ▁OUT
- PAUSE
- EST
- ARRIVAL
- ▁CANCEL
- ▁MI
- ▁OFF
- ▁FIFTEEN
- POINT
- ▁SNOW
- NA
- EL
- ▁EVENTS
- ▁CA
- ▁SUNDAY
- ▁LEAVE
- TRACK
- ▁SEND
- ▁DELETE
- ▁APPOINTMENT
- ▁BO
- RDINAL
- ▁MAKE
- ▁NEAR
- ▁BEFORE
- GE
- ▁HOME
- RELATION
- ▁V
- FR
- ▁THURSDAY
- ▁LAST
- DIRECTIONS
- ▁WEDNESDAY
- ▁START
- ▁FORECAST
- ▁YORK
- ▁RIGHT
- UM
- ▁WITH
- USE
- ▁MEETING
- UT
- LI
- ▁CHANGE
- ▁CAR
- GENRE
- ATION
- X
- ▁PICK
- ▁WANT
- ▁NIGHT
- SKIP
- ▁DE
- ▁RO
- ▁ABOUT
- MAP
- CO
- MA
- ▁HOUSE
- ▁HOT
- ▁PARTY
- ▁WA
- UNIT
- ▁HERE
- ▁SU
- ▁AFTERNOON
- ▁MUCH
- ▁MOM
- ▁TEMPERATURE
- EQUENC
- ▁ADD
- ▁SAN
- ▁HER
- ▁CONCERTS
- ▁CHRISTMAS
- ▁DINNER
- ▁MAR
- LAND
- ▁HOURS
- ▁CURRENT
- ▁TRACK
- ▁SOME
- ▁CITY
- ▁FORTY
- ATE
- ▁ROUTE
- SNOOZE
- ▁TEXT
- WORK
- ▁COLD
- RELATED
- ▁OR
- ▁NO
- Q
- ▁WAY
- WAY
- ▁MANY
- ▁BIRTHDAY
- ▁MINUTE
- ▁PLAYLIST
- ▁NOON
- ▁ROAD
- TITLE
- PATH
- ▁ASK
- NAVIGATION
- ▁LEFT
- ▁ALBUM
- ▁TURN
- ▁LATE
- ▁ELEVEN
- NEW
- ▁CELSIUS
- ▁BUY
- AVOID
- LOW
- NCE
- SEARCH
- ▁GAME
- ▁STOP
- ▁JO
- ▁FIRST
- ▁SHE
- ▁DOCTOR
- ▁BU
- PERIOD
- ▁WAKE
- CONDITION
- ▁EVENING
- RADIUS
- MODIFIE
- ▁REPEAT
- ▁SECOND
- ▁CONCERT
- ▁ANGELES
- ▁DOWNTOWN
- ▁UMBRELLA
- TEMPERATURE
- ASH
- ▁YEAR
- GROUP
- ▁DRIVING
- ▁GIVE
- ▁HUNDRED
- ▁HO
- ▁MILES
- PLAYLIST
- ADD
- RETRIEV
- ▁TWELVE
- EAD
- ▁CLASS
- ▁FREE
- PORT
- VILLE
- ▁BETWEEN
- ▁KNOW
- ▁AROUND
- ▁SCHOOL
- ▁NINETY
- PROVIDER
- SILENCE
- RESUME
- ▁LET
- TION
- ▁AUGUST
- ▁HAPPENING
- ▁AFTER
- ▁FAHRENHEIT
- ▁EX
- ▁VIDEO
- ROAD
- ▁PARK
- ▁CHICAGO
- ▁DAILY
- ▁CHECK
- ▁BEACH
- ▁WHERE
- ▁JUNE
- ▁STREET
- ▁FESTIVAL
- ▁FLORIDA
- ▁JOHN
- ▁HAS
- ▁SPOTIFY
- ▁BILL
- RESTART
- ▁HIGHWAY
- ▁SEATTLE
- J
- ▁LUNCH
- ▁LOOK
- ▁FRIEND
- ▁COMING
- ▁ALERT
- IGHT
- ▁PANDORA
- ▁HEAVY
- ▁KIDS
- ▁MOVIE
- ▁SOUTH
- REACT
- ▁CONSTRUCTION
- PREVIOUS
- ▁ORLANDO
- ▁OVER
- ▁MIAMI
- REACTION
- ▁ATLANTA
- ▁ACCIDENT
- ▁COUNTRY
- ▁NORTH
- ▁LIGHT
- RADIO
- ▁READ
- ▁FAMILY
- ▁AIRPORT
- ▁EXPECT
- ▁DEGREE
- ▁PRO
- ▁PARTIES
- ▁FIFTY
- ▁HIGH
- ▁PLAN
- ▁FOOD
- ▁WARM
- ▁SUNNY
- ▁VEGAS
- ▁HOLIDAY
- ▁SCHEDULE
- ▁STORM
- ▁FIFTH
- ▁BOSTON
- ▁FRANCISCO
- ▁LONDON
- ATTENDEE
- ▁JULY
- ▁WALK
- ▁COMMUTE
- ▁CLEAN
- ▁DENTIST
- TOWN
- ▁AGAIN
- ▁DALLAS
- ▁PORTLAND
- ▁SEPTEMBER
- ▁ARRIVE
- ▁SISTER
- ▁HOUSTON
- Ã
- É
- Í
- '*'
- Á
- Ç
- Ó
- ']'
- '['
- Ú
- Ü
- <sos/eos>
transcript_token_list: null
two_pass: false
pre_postencoder_norm: false
init: null
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true
joint_net_conf: null
use_preprocessor: true
token_type: bpe
bpemodel: data/en_token_list/bpe_unigram500/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: s3prl
frontend_conf:
    frontend_conf:
        upstream: hubert_large_ll60k
    download_dir: ./hub
    multilayer_feature: true
    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: 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: 80
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: conv2d2
    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: 31
postencoder: null
postencoder_conf: {}
deliberationencoder: null
deliberationencoder_conf: {}
decoder: transformer
decoder_conf:
    attention_heads: 8
    linear_units: 2048
    num_blocks: 6
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
decoder2: null
decoder2_conf: {}
postdecoder: null
postdecoder_conf: {}
required:
- output_dir
- token_list
version: '202205'
distributed: true

</details>

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}
}