voxcelebs12_xvector_mel

Usage

import librosa
from espnet2.bin.spk_inference import Speech2Embedding

speech2embedding = Speech2Embedding.from_pretrained(model_tag="espnet/voxcelebs12_xvector_mel")
# the speaker models are trained on 16 kHz mono; librosa gives that from
# whatever the file holds
speech, rate = librosa.load("audio.wav", sr=16000, mono=True)
embedding = speech2embedding(speech)  # (1, embedding_dim)

ESPnet2 SPK model

espnet/voxcelebs12_xvector_mel

This model was trained by Jungjee using voxceleb 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 e5da124138fc58708fcdda03fd8d4e02fe5d7c65
pip install -e .
cd egs2/voxceleb/spk1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/voxcelebs12_xvector_mel

<!-- Generated by scripts/utils/show_spk_result.py -->

RESULTS

Environments

date: 2023-12-27 19:06:52.486861

• python version: 3.9.16 (main, Mar 8 2023, 14:00:05) [GCC 11.2.0]
• espnet version: 202310
• pytorch version: 2.0.1

SPK config

<details><summary>expand</summary>

config: conf/tuning/train_xvector_Vox12_amp_subcentertopk.yaml
print_config: false
log_level: INFO
drop_last_iter: true
dry_run: false
iterator_type: category
valid_iterator_type: sequence
output_dir: exp/spk_train_xvector_Vox12_amp_subcentertopk_raw_sp
ngpu: 1
seed: 0
num_workers: 6
num_att_plot: 0
dist_backend: nccl
dist_init_method: env://
dist_world_size: 4
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 60911
dist_launcher: null
multiprocessing_distributed: true
unused_parameters: false
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: true
cudnn_deterministic: false
collect_stats: false
write_collected_feats: false
max_epoch: 40
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - eer
    - min
keep_nbest_models: 3
nbest_averaging_interval: 0
grad_clip: 9999
grad_clip_type: 2.0
grad_noise: false
accum_grad: 1
no_forward_run: false
resume: true
train_dtype: float32
use_amp: true
log_interval: 100
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_lora: false
save_lora_only: true
lora_conf: {}
pretrain_path: null
init_param: []
ignore_init_mismatch: false
freeze_param: []
num_iters_per_epoch: null
batch_size: 512
valid_batch_size: 40
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/spk_stats_16k_sp/train/speech_shape
valid_shape_file:
- exp/spk_stats_16k_sp/valid/speech_shape
batch_type: folded
valid_batch_type: null
fold_length:
- 120000
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
train_data_path_and_name_and_type:
-   - dump/raw/voxceleb12_devs_sp/wav.scp
    - speech
    - sound
-   - dump/raw/voxceleb12_devs_sp/utt2spk
    - spk_labels
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/voxceleb1_test/trial.scp
    - speech
    - sound
-   - dump/raw/voxceleb1_test/trial2.scp
    - speech2
    - sound
-   - dump/raw/voxceleb1_test/trial_label
    - spk_labels
    - text
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: adam
optim_conf:
    lr: 0.001
    weight_decay: 5.0e-05
    amsgrad: false
scheduler: cosineannealingwarmuprestarts
scheduler_conf:
    first_cycle_steps: 71280
    cycle_mult: 1.0
    max_lr: 0.001
    min_lr: 5.0e-06
    warmup_steps: 1000
    gamma: 0.75
init: null
use_preprocessor: true
input_size: null
target_duration: 3.0
spk2utt: dump/raw/voxceleb12_devs_sp/spk2utt
spk_num: 21615
sample_rate: 16000
num_eval: 10
rir_scp: ''
model_conf:
    extract_feats_in_collect_stats: false
frontend: melspec_torch
frontend_conf:
    preemp: true
    n_fft: 512
    log: true
    win_length: 400
    hop_length: 160
    n_mels: 80
    normalize: mn
specaug: null
specaug_conf: {}
normalize: null
normalize_conf: {}
encoder: xvector
encoder_conf:
    ndim: 512
    output_size: 1500
pooling: stats
pooling_conf: {}
projector: xvector
projector_conf:
    output_size: 512
preprocessor: spk
preprocessor_conf:
    target_duration: 3.0
    sample_rate: 16000
    num_eval: 5
    noise_apply_prob: 0.5
    noise_info:
    -   - 1.0
        - dump/raw/musan_speech.scp
        -   - 4
            - 7
        -   - 13
            - 20
    -   - 1.0
        - dump/raw/musan_noise.scp
        -   - 1
            - 1
        -   - 0
            - 15
    -   - 1.0
        - dump/raw/musan_music.scp
        -   - 1
            - 1
        -   - 5
            - 15
    rir_apply_prob: 0.5
    rir_scp: dump/raw/rirs.scp
loss: aamsoftmax_sc_topk
loss_conf:
    margin: 0.3
    scale: 30
    K: 3
    mp: 0.06
    k_top: 5
required:
- output_dir
version: '202310'
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}
}