hubert_large_gs_16_librilight60k

ESPnet2 SSL model

espnet/hubert_large_gs_16_librilight60k

This model was trained by wanchichen using librispeech recipe in espnet.

Demo: How to use in ESPnet2

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

cd espnet

pip install -e .
cd egs2/librispeech/ssl1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/hubert_large_gs_16_librilight60k

SSL config

<details><summary>expand</summary>

config: conf/tuning/train_ssl_torchaudiohubert_large_960h_pretrain_it2_bins.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/hubert_iter1_train_ssl_torchaudiohubert_large_960h_pretrain_it2_bins_raw
ngpu: 1
seed: 0
num_workers: 16
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 8
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 55415
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: 190
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - loss
    - min
keep_nbest_models: 10
nbest_averaging_interval: 0
grad_clip: 5.0
grad_clip_type: 2.0
grad_noise: false
accum_grad: 16
no_forward_run: false
resume: true
train_dtype: float32
use_amp: true
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: []
num_iters_per_epoch: 40000
batch_size: 20
valid_batch_size: null
batch_bins: 100000
valid_batch_bins: null
train_shape_file:
- exp/hubert_iter1_stats_raw/splits16/speech_shape
- exp/hubert_iter1_stats_raw/splits16/text_shape.word
valid_shape_file:
- exp/hubert_iter1_stats_raw/valid/speech_shape
- exp/hubert_iter1_stats_raw/valid/text_shape.word
batch_type: numel
valid_batch_type: null
fold_length:
- 80000
- 400
sort_in_batch: descending
sort_batch: descending
multiple_iterator: true
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
train_data_path_and_name_and_type:
-   - exp/hubert_iter1_stats_raw/splits16/wav.scp
    - speech
    - sound
-   - exp/hubert_iter1_stats_raw/splits16/text.km.kmeans_iter1_hubert_train_60k_portion0.1_gigaspeech
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dev/wav.scp
    - speech
    - sound
-   - dump/raw/dev/text.km.kmeans_iter1_hubert_train_60k_portion0.1_gigaspeech
    - 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: 32000
token_list:
- '6'
- '185'
- '233'
- '206'
- '200'
- '47'
- '129'
- '362'
- '436'
- '50'
- '30'
- '137'
- '39'
- '126'
- '81'
- '78'
- '444'
- '439'
- '230'
- '33'
- '14'
- '61'
- '450'
- '239'
- '293'
- '161'
- '410'
- '355'
- '262'
- '475'
- '338'
- '201'
- '242'
- '318'
- '159'
- '56'
- '190'
- '21'
- '259'
- '458'
- '187'
- '268'
- '7'
- '398'
- '67'
- '11'
- '455'
- '241'
- '4'
- '261'
- '412'
- '388'
- '402'
- '414'
- '180'
- '41'
- '198'
- '100'
- '459'
- '96'
- '235'
- '267'
- '203'
- '189'
- '60'
- '160'
- '64'
- '300'
- '295'
- '473'
- '37'
- '130'
- '77'
- '424'
- '369'
- '217'
- '366'
- '3'
- '82'
- '149'
- '330'
- '79'
- '351'
- '223'
- '446'
- '269'
- '148'
- '186'
- '110'
- '62'
- '471'
- '316'
- '433'
- '127'
- '354'
- '243'
- '457'
- '240'
- '375'
- '46'
- '40'
- '339'
- '224'
- '183'
- '179'
- '357'
- '430'
- '83'
- '49'
- '154'
- '237'
- '460'
- '353'
- '289'
- '92'
- '109'
- '311'
- '71'
- '391'
- '406'
- '43'
- '73'
- '418'
- '437'
- '250'
- '463'
- '120'
- '346'
- '146'
- '454'
- '211'
- '274'
- '167'
- '345'
- '10'
- '68'
- '348'
- '244'
- '102'
- '474'
- '192'
- '144'
- '112'
- '25'
- '449'
- '308'
- '405'
- '48'
- '212'
- '205'
- '124'
- '153'
- '9'
- '5'
- '258'
- '306'
- '80'
- '394'
- '328'
- '208'
- '166'
- '36'
- '352'
- '18'
- '397'
- '66'
- '31'
- '16'
- '426'
- '332'
- '23'
- '281'
- '215'
- '88'
- '171'
- '221'
- '184'
- '202'
- '470'
- '247'
- '38'
- '389'
- '315'
- '197'
- '349'
- '304'
- '393'
- '380'
- '132'
- '456'
- '367'
- '479'
- '360'
- '123'
- '162'
- '365'
- '337'
- '467'
- '234'
- '364'
- '376'
- '173'
- '478'
- '425'
- '218'
- '297'
- '469'
- '282'
- '298'
- '451'
- '20'
- '117'
- '52'
- '113'
- '165'
- '280'
- '292'
- '226'
- '104'
- '55'
- '145'
- '286'
- '86'
- '294'
- '15'
- '216'
- '279'
- '275'
- '253'
- '312'
- '378'
- '287'
- '76'
- '168'
- '116'
- '368'
- '396'
- '336'
- '290'
- '53'
- '103'
- '0'
- '411'
- '228'
- '408'
- '285'
- '151'
- '325'
- '193'
- '428'
- '401'
- '320'
- '182'
- '480'
- '264'
- '383'
- '114'
- '115'
- '374'
- '141'
- '22'
- '466'
- '384'
- '174'
- '59'
- '326'
- '105'
- '232'
- '464'
- '251'
- '24'
- '172'
- '150'
- '299'
- '89'
- '344'
- '427'
- '333'
- '434'
- '107'
- '291'
- '194'
- '497'
- '452'
- '317'
- '254'
- '213'
- '499'
- '483'
- '432'
- '95'
- '321'
- '111'
- '8'
- '175'
- '277'
- '65'
- '342'
- '382'
- '301'
- '45'
- '443'
- '63'
- '93'
- '489'
- '74'
- '387'
- '370'
- '340'
- '358'
- '220'
- '429'
- '2'
- '331'
- '181'
- '32'
- '324'
- '191'
- '238'
- '313'
- '157'
- '91'
- '101'
- '118'
- '350'
- '356'
- '486'
- '188'
- '142'
- '419'
- '195'
- '164'
- '487'
- '255'
- '323'
- '222'
- '35'
- '245'
- '359'
- '249'
- '98'
- '271'
- '231'
- '125'
- '29'
- '34'
- '119'
- '134'
- '284'
- '309'
- '409'
- '422'
- '147'
- '484'
- '462'
- '390'
- '440'
- '283'
- '84'
- '108'
- '139'
- '170'
- '303'
- '371'
- '381'
- '278'
- '329'
- '28'
- '87'
- '403'
- '256'
- '441'
- '334'
- '12'
- '260'
- '265'
- '69'
- '122'
- '488'
- '99'
- '42'
- '302'
- '97'
- '70'
- '152'
- '177'
- '138'
- '296'
- '51'
- '491'
- '199'
- '176'
- '204'
- '169'
- '386'
- '494'
- '400'
- '341'
- '229'
- '273'
- '485'
- '135'
- '227'
- '54'
- '314'
- '343'
- '477'
- '465'
- '482'
- '257'
- '435'
- '423'
- '121'
- '496'
- '448'
- '453'
- '85'
- '57'
- '276'
- '210'
- '272'
- '236'
- '407'
- '445'
- '90'
- '266'
- '490'
- '307'
- '155'
- '136'
- '19'
- '319'
- '498'
- '163'
- '75'
- '442'
- '495'
- '421'
- '209'
- '361'
- '156'
- '395'
- '472'
- '415'
- '347'
- '252'
- '468'
- '476'
- '106'
- '143'
- '263'
- '373'
- '327'
- '322'
- '399'
- '404'
- '13'
- '288'
- '207'
- '58'
- '481'
- '131'
- '385'
- '447'
- '219'
- '438'
- '461'
- '416'
- '246'
- '417'
- '26'
- '158'
- '431'
- '270'
- '128'
- '413'
- '310'
- '140'
- '17'
- '392'
- '44'
- '27'
- '214'
- '377'
- '305'
- '72'
- '420'
- '133'
- '363'
- '379'
- '94'
- '225'
- '335'
- '493'
- '492'
- '372'
- '196'
- '248'
- '178'
- '1'
- <unk>
- <sos/eos>
init: null
collate_fn_conf:
    label_downsampling: 1
    pad: false
    rand_crop: true
input_size: 1
num_classes: 500
use_preprocessor: true
token_type: word
bpemodel: null
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'
pred_masked_weight: 1.0
pred_nomask_weight: 0.0
loss_weights: 0.0
frontend: null
frontend_conf: {}
specaug: null
specaug_conf: {}
normalize: null
normalize_conf: {}
preencoder: null
preencoder_conf: {}
encoder: torchaudio_hubert
encoder_conf:
    encoder_projection_dropout: 0.0
    encoder_attention_dropout: 0.0
    encoder_ff_interm_dropout: 0.0
    encoder_dropout: 0.0
    encoder_layer_drop: 0.0
    extractor_mode: layer_norm
    encoder_embed_dim: 1024
    encoder_num_layers: 24
    encoder_num_heads: 16
    encoder_ff_interm_features: 4096
    encoder_layer_norm_first: true
    normalize_feats: true
    final_dim: 768
model: torchaudio
model_conf: {}
required:
- output_dir
- token_list
version: '202301'
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}
}