aceopencpop_svs_visinger2_40singer_pretrain

SingingSDS: A Singing-Capable Spoken Dialogue System for Conversational Roleplay Applications

This repository contains the espnet/aceopencpop_svs_visinger2_40singer_pretrain model, an ESPnet2 Singing Voice Synthesis (SVS) component used within the broader SingingSDS: A Singing-Capable Spoken Dialogue System for Conversational Roleplay Applications project.

• Paper: SingingSDS: A Singing-Capable Spoken Dialogue System for Conversational Roleplay Applications
• Code (SingingSDS GitHub): https://github.com/SingingSDS/SingingSDS
• Demo (Hugging Face Space): https://huggingface.co/spaces/espnet/SingingSDS

ESPnet2 SVS model

espnet/aceopencpop_svs_visinger2_40singer_pretrain

This model was trained by ftshijt using acesinger 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 ba4880118d5249e5dd92e89d107280a0d4f317e8
pip install -e .
cd egs2/acesinger/svs1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/aceopencpop_svs_visinger2_40singer_pretrain

SVS config

<details><summary>expand</summary>

config: conf/tuning/train_visinger2_40singer.yaml
print_config: false
log_level: INFO
drop_last_iter: false
dry_run: false
iterator_type: sequence
valid_iterator_type: null
output_dir: exp/svs_train_visinger2_40singer_raw_phn_None_zh
ngpu: 1
seed: 777
num_workers: 0
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: true
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: false
collect_stats: false
write_collected_feats: false
max_epoch: 500
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - train
    - total_count
    - max
keep_nbest_models: 10
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_lora: false
save_lora_only: true
lora_conf: {}
pretrain_path: null
init_param: []
ignore_init_mismatch: false
freeze_param: []
num_iters_per_epoch: 1000
batch_size: 8
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/svs_stats_raw_phn_None_zh/train/text_shape.phn
- exp/svs_stats_raw_phn_None_zh/train/singing_shape
valid_shape_file:
- exp/svs_stats_raw_phn_None_zh/valid/text_shape.phn
- exp/svs_stats_raw_phn_None_zh/valid/singing_shape
batch_type: sorted
valid_batch_type: null
fold_length:
- 150
- 409600
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/tr_no_dev/text
    - text
    - text
-   - dump/raw/tr_no_dev/wav.scp
    - singing
    - sound
-   - dump/raw/tr_no_dev/label
    - label
    - duration
-   - dump/raw/tr_no_dev/score.scp
    - score
    - score
-   - exp/svs_stats_raw_phn_None_zh/train/collect_feats/pitch.scp
    - pitch
    - npy
-   - exp/svs_stats_raw_phn_None_zh/train/collect_feats/feats.scp
    - feats
    - npy
-   - dump/raw/tr_no_dev/utt2sid
    - sids
    - text_int
valid_data_path_and_name_and_type:
-   - dump/raw/dev/text
    - text
    - text
-   - dump/raw/dev/wav.scp
    - singing
    - sound
-   - dump/raw/dev/label
    - label
    - duration
-   - dump/raw/dev/score.scp
    - score
    - score
-   - exp/svs_stats_raw_phn_None_zh/valid/collect_feats/pitch.scp
    - pitch
    - npy
-   - exp/svs_stats_raw_phn_None_zh/valid/collect_feats/feats.scp
    - feats
    - npy
-   - dump/raw/dev/utt2sid
    - sids
    - text_int
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: 0.0002
    betas:
    - 0.8
    - 0.99
    eps: 1.0e-09
    weight_decay: 0.0
scheduler: exponentiallr
scheduler_conf:
    gamma: 0.998
optim2: adamw
optim2_conf:
    lr: 0.0002
    betas:
    - 0.8
    - 0.99
    eps: 1.0e-09
    weight_decay: 0.0
scheduler2: exponentiallr
scheduler2_conf:
    gamma: 0.998
generator_first: true
token_list:
- <blank>
- <unk>
- SP
- i
- AP
- e
- d
- y
- w
- sh
- ai
- n
- x
- j
- u
- ian
- l
- h
- b
- o
- zh
- ou
- an
- m
- q
- z
- en
- g
- ing
- ei
- ao
- uo
- ang
- eng
- t
- ong
- a
- ui
- f
- k
- r
- ch
- v
- iang
- in
- iao
- ie
- iu
- c
- s
- van
- p
- ve
- uan
- uang
- ia
- ua
- uai
- un
- er
- vn
- iong
- <sos/eos>
odim: null
model_conf: {}
use_preprocessor: true
token_type: phn
bpemodel: null
non_linguistic_symbols: null
cleaner: null
g2p: null
fs: 44100
score_feats_extract: syllable_score_feats
score_feats_extract_conf:
    fs: 44100
    n_fft: 2048
    win_length: 2048
    hop_length: 512
feats_extract: fbank
feats_extract_conf:
    n_fft: 2048
    hop_length: 512
    win_length: 2048
    fs: 44100
    fmin: 80
    fmax: 7600
    n_mels: 80
normalize: global_mvn
normalize_conf:
    stats_file: exp/svs_stats_raw_phn_None_zh/train/feats_stats.npz
svs: vits
svs_conf:
    generator_type: visinger2
    vocoder_generator_type: visinger2
    generator_params:
        hidden_channels: 192
        spks: 40
        global_channels: 256
        segment_size: 20
        text_encoder_attention_heads: 2
        text_encoder_ffn_expand: 4
        text_encoder_blocks: 6
        text_encoder_positionwise_layer_type: conv1d
        text_encoder_positionwise_conv_kernel_size: 3
        text_encoder_positional_encoding_layer_type: rel_pos
        text_encoder_self_attention_layer_type: rel_selfattn
        text_encoder_activation_type: swish
        text_encoder_normalize_before: true
        text_encoder_dropout_rate: 0.1
        text_encoder_positional_dropout_rate: 0.0
        text_encoder_attention_dropout_rate: 0.1
        use_macaron_style_in_text_encoder: true
        use_conformer_conv_in_text_encoder: false
        text_encoder_conformer_kernel_size: -1
        decoder_kernel_size: 7
        decoder_channels: 256
        decoder_upsample_scales:
        - 8
        - 8
        - 4
        - 2
        decoder_upsample_kernel_sizes:
        - 16
        - 16
        - 8
        - 4
        n_harmonic: 64
        decoder_resblock_kernel_sizes:
        - 3
        - 7
        - 11
        decoder_resblock_dilations:
        -   - 1
            - 3
            - 5
        -   - 1
            - 3
            - 5
        -   - 1
            - 3
            - 5
        use_weight_norm_in_decoder: true
        posterior_encoder_kernel_size: 3
        posterior_encoder_layers: 8
        posterior_encoder_stacks: 1
        posterior_encoder_base_dilation: 1
        posterior_encoder_dropout_rate: 0.0
        use_weight_norm_in_posterior_encoder: true
        flow_flows: -1
        flow_kernel_size: 5
        flow_base_dilation: 1
        flow_layers: 4
        flow_dropout_rate: 0.0
        use_weight_norm_in_flow: true
        use_only_mean_in_flow: true
        use_phoneme_predictor: false
        vocabs: 63
        aux_channels: 80
        generator_type: visinger2
        vocoder_generator_type: visinger2
        fs: 44100
        hop_length: 512
        win_length: 2048
        n_fft: 2048
    discriminator_type: visinger2
    discriminator_params:
        scales: 1
        scale_downsample_pooling: AvgPool1d
        scale_downsample_pooling_params:
            kernel_size: 4
            stride: 2
            padding: 2
        scale_discriminator_params:
            in_channels: 1
            out_channels: 1
            kernel_sizes:
            - 15
            - 41
            - 5
            - 3
            channels: 128
            max_downsample_channels: 1024
            max_groups: 256
            bias: true
            downsample_scales:
            - 4
            - 4
            - 4
            - 4
            nonlinear_activation: LeakyReLU
            nonlinear_activation_params:
                negative_slope: 0.1
            use_weight_norm: true
            use_spectral_norm: false
        follow_official_norm: false
        periods:
        - 2
        - 3
        - 5
        - 7
        - 11
        period_discriminator_params:
            in_channels: 1
            out_channels: 1
            kernel_sizes:
            - 5
            - 3
            channels: 32
            downsample_scales:
            - 3
            - 3
            - 3
            - 3
            - 1
            max_downsample_channels: 1024
            bias: true
            nonlinear_activation: LeakyReLU
            nonlinear_activation_params:
                negative_slope: 0.1
            use_weight_norm: true
            use_spectral_norm: false
        multi_freq_disc_params:
            hidden_channels:
            - 256
            - 256
            - 256
            - 256
            - 256
            domain: double
            mel_scale: true
            divisors:
            - 32
            - 16
            - 8
            - 4
            - 2
            - 1
            - 1
            strides:
            - 1
            - 2
            - 1
            - 2
            - 1
            - 2
            - 1
            sample_rate: 44100
            hop_lengths:
            - 110
            - 220
            - 330
            - 441
            - 551
            - 661
    generator_adv_loss_params:
        average_by_discriminators: false
        loss_type: mse
    discriminator_adv_loss_params:
        average_by_discriminators: false
        loss_type: mse
    feat_match_loss_params:
        average_by_discriminators: false
        average_by_layers: false
        include_final_outputs: true
    mel_loss_params:
        fs: 44100
        n_fft: 2048
        hop_length: 512
        win_length: 2048
        window: hann
        n_mels: 80
        fmin: 0
        fmax: 22050
        log_base: null
    lambda_adv: 1.0
    lambda_mel: 45.0
    lambda_feat_match: 2.0
    lambda_dur: 0.1
    lambda_pitch: 10.0
    lambda_phoneme: 1.0
    lambda_kl: 1.0
    sampling_rate: 44100
    cache_generator_outputs: true
pitch_extract: dio
pitch_extract_conf:
    use_token_averaged_f0: false
    use_log_f0: false
    fs: 44100
    n_fft: 2048
    hop_length: 512
    f0max: 800
    f0min: 80
pitch_normalize: null
pitch_normalize_conf:
    stats_file: exp/svs_stats_raw_phn_None_zh/train/pitch_stats.npz
ying_extract: null
ying_extract_conf: {}
energy_extract: null
energy_extract_conf: {}
energy_normalize: null
energy_normalize_conf: {}
required:
- output_dir
- token_list
version: '202310'
distributed: false

</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}
}






@inproceedings{shi22d_interspeech,
  author={Jiatong Shi and Shuai Guo and Tao Qian and Tomoki Hayashi and Yuning Wu and Fangzheng Xu and Xuankai Chang and Huazhe Li and Peter Wu and Shinji Watanabe and Qin Jin},
  title={{Muskits: an End-to-end Music Processing Toolkit for Singing Voice Synthesis}},
  year=2022,
  booktitle={Proc. Interspeech 2022},
  pages={4277--4281},
  doi={10.21437/Interspeech.2022-10039}
}

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}
}