Upload 12 files

configs/astral_quantization/default_2048.yml

@@ -0,0 +1,40 @@
+_target_: modules.astral_quantization.default_model.AstralQuantizer
+tokenizer_name: "openai/whisper-small"
+ssl_model_name: "facebook/hubert-large-ll60k"
+ssl_output_layer: 18
+encoder:
+  _target_: modules.astral_quantization.convnext.ConvNeXtV2Stage
+  dim: 512
+  num_blocks: 12
+  intermediate_dim: 1536
+  dilation: 1
+  input_dim: 1024
+quantizer:
+  _target_: modules.astral_quantization.bsq.BinarySphericalQuantize
+  codebook_size: 2048  # codebook size, must be a power of 2
+  dim: 512
+  entropy_loss_weight: 0.1
+  diversity_gamma: 1.0
+  spherical: True
+  enable_entropy_loss: True
+  soft_entropy_loss: True
+decoder:
+  _target_: modules.astral_quantization.convnext.ConvNeXtV2Stage
+  dim: 512
+  num_blocks: 12
+  intermediate_dim: 1536
+  dilation: 1
+  output_dim: 1024
+  gin_channels: 192
+asr_decoder:
+  _target_: modules.astral_quantization.asr_decoder.ASRDecoder
+  hidden_dim: 768
+  num_heads: 12
+  depth: 12
+  block_size: 4096
+  in_channels: 512
+  n_vocab: 51866
+  bos_id: 50528
+  eos_id: 50527
+  dropout_rate: 0.0
+  attn_dropout_rate: 0.0

configs/astral_quantization/default_32.yml

@@ -0,0 +1,40 @@
+_target_: default_model.AstralQuantizer
+tokenizer_name: "openai/whisper-small"
+ssl_model_name: "facebook/hubert-large-ll60k"
+ssl_output_layer: 18
+encoder:
+  _target_: modules.convnext.ConvNeXtV2Stage
+  dim: 512
+  num_blocks: 12
+  intermediate_dim: 1536
+  dilation: 1
+  input_dim: 1024
+quantizer:
+  _target_: modules.bsq.BinarySphericalQuantize
+  codebook_size: 32  # codebook size, must be a power of 2
+  dim: 512
+  entropy_loss_weight: 0.1
+  diversity_gamma: 1.0
+  spherical: True
+  enable_entropy_loss: True
+  soft_entropy_loss: True
+decoder:
+  _target_: modules.convnext.ConvNeXtV2Stage
+  dim: 512
+  num_blocks: 12
+  intermediate_dim: 1536
+  dilation: 1
+  output_dim: 1024
+  gin_channels: 192
+asr_decoder:
+  _target_: modules.asr_decoder.ASRDecoder
+  hidden_dim: 768
+  num_heads: 12
+  depth: 12
+  block_size: 4096
+  in_channels: 512
+  n_vocab: 51866
+  bos_id: 50528
+  eos_id: 50527
+  dropout_rate: 0.0
+  attn_dropout_rate: 0.0

configs/config.json

@@ -0,0 +1 @@
+{"reference_audio_path": "D:/FAcodec/test_waves/kobe_0.wav", "sg_hostapi": "MME", "sg_wasapi_exclusive": false, "sg_input_device": "\u9ea6\u514b\u98ce (Razer BlackShark V2 HS 2.4", "sg_output_device": "\u626c\u58f0\u5668 (Razer BlackShark V2 HS 2.4", "sr_type": "sr_model", "diffusion_steps": 10.0, "inference_cfg_rate": 0.0, "max_prompt_length": 3.0, "block_time": 0.7, "crossfade_length": 0.04, "extra_time": 0.5, "extra_time_right": 0.02}

configs/config_dit_mel_seed.yml

@@ -0,0 +1,79 @@
+log_dir: "./runs/run_dit_mel_seed"
+save_freq: 1
+log_interval: 10
+save_interval: 1000
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 4
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: ""
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+F0_path: "modules/JDC/bst.t7"
+
+preprocess_params:
+  sr: 22050
+  spect_params:
+    n_fft: 1024
+    win_length: 1024
+    hop_length: 256
+    n_mels: 80
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l2" # l1 or l2
+
+  speech_tokenizer:
+    path: "speech_tokenizer_v1.onnx"
+
+  style_encoder:
+    dim: 192
+    campplus_path: "campplus_cn_common.bin"
+
+  DAC:
+    encoder_dim: 64
+    encoder_rates: [2, 5, 5, 6]
+    decoder_dim: 1536
+    decoder_rates: [ 6, 5, 5, 2 ]
+    sr: 24000
+
+  length_regulator:
+    channels: 768
+    is_discrete: true
+    content_codebook_size: 4096
+    in_frame_rate: 50
+    out_frame_rate: 80
+    sampling_ratios: [1, 1, 1, 1]
+
+  DiT:
+    hidden_dim: 768
+    num_heads: 12
+    depth: 12
+    class_dropout_prob: 0.1
+    block_size: 4096
+    in_channels: 80
+    style_condition: true
+    final_layer_type: 'wavenet'
+    target: 'mel' # mel or codec
+    content_dim: 768
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: false
+    n_f0_bins: 512
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: true
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+
+  wavenet:
+    hidden_dim: 768
+    num_layers: 8
+    kernel_size: 5
+    dilation_rate: 1
+    p_dropout: 0.2
+    style_condition: true
+
+loss_params:
+  base_lr: 0.0001

configs/config_dit_mel_seed_facodec_small.yml

@@ -0,0 +1,97 @@
+log_dir: "./runs/run_dit_mel_seed_facodec_small"
+save_freq: 1
+log_interval: 10
+save_interval: 1000
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 2
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: ""
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+F0_path: "modules/JDC/bst.t7"
+
+data_params:
+  train_data: "./data/train.txt"
+  val_data: "./data/val.txt"
+  root_path: "./data/"
+
+preprocess_params:
+  sr: 22050
+  spect_params:
+    n_fft: 1024
+    win_length: 1024
+    hop_length: 256
+    n_mels: 80
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l1" # l1 or l2
+
+  speech_tokenizer:
+    type: 'facodec' # facodec or cosyvoice
+    path: "checkpoints/speech_tokenizer_v1.onnx"
+
+  style_encoder:
+    dim: 192
+    campplus_path: "checkpoints/campplus_cn_common.bin"
+
+  DAC:
+    encoder_dim: 64
+    encoder_rates: [2, 5, 5, 6]
+    decoder_dim: 1536
+    decoder_rates: [ 6, 5, 5, 2 ]
+    sr: 24000
+
+  length_regulator:
+    channels: 512
+    is_discrete: true
+    content_codebook_size: 1024
+    in_frame_rate: 80
+    out_frame_rate: 80
+    sampling_ratios: [1, 1, 1, 1]
+    token_dropout_prob: 0.3 # probability of performing token dropout
+    token_dropout_range: 1.0 # maximum percentage of tokens to drop out
+    n_codebooks: 3
+    quantizer_dropout: 0.5
+    f0_condition: false
+    n_f0_bins: 512
+
+  DiT:
+    hidden_dim: 512
+    num_heads: 8
+    depth: 13
+    class_dropout_prob: 0.1
+    block_size: 8192
+    in_channels: 80
+    style_condition: true
+    final_layer_type: 'wavenet'
+    target: 'mel' # mel or codec
+    content_dim: 512
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: true
+    n_f0_bins: 512
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: true
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+    time_as_token: false
+    style_as_token: false
+    uvit_skip_connection: true
+    add_resblock_in_transformer: false
+
+  wavenet:
+    hidden_dim: 512
+    num_layers: 8
+    kernel_size: 5
+    dilation_rate: 1
+    p_dropout: 0.2
+    style_condition: true
+
+loss_params:
+  base_lr: 0.0001
+  lambda_mel: 45
+  lambda_kl: 1.0

configs/config_dit_mel_seed_wavenet.yml

@@ -0,0 +1,79 @@
+log_dir: "./runs/run_dit_mel_seed"
+save_freq: 1
+log_interval: 10
+save_interval: 1000
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 4
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: ""
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+F0_path: "modules/JDC/bst.t7"
+
+preprocess_params:
+  sr: 22050
+  spect_params:
+    n_fft: 1024
+    win_length: 1024
+    hop_length: 256
+    n_mels: 80
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l2" # l1 or l2
+
+  speech_tokenizer:
+    path: "checkpoints/speech_tokenizer_v1.onnx"
+
+  style_encoder:
+    dim: 192
+    campplus_path: "campplus_cn_common.bin"
+
+  DAC:
+    encoder_dim: 64
+    encoder_rates: [2, 5, 5, 6]
+    decoder_dim: 1536
+    decoder_rates: [ 6, 5, 5, 2 ]
+    sr: 24000
+
+  length_regulator:
+    channels: 768
+    is_discrete: true
+    content_codebook_size: 4096
+    in_frame_rate: 50
+    out_frame_rate: 80
+    sampling_ratios: [1, 1, 1, 1]
+
+  DiT:
+    hidden_dim: 768
+    num_heads: 12
+    depth: 12
+    class_dropout_prob: 0.1
+    block_size: 8192
+    in_channels: 80
+    style_condition: true
+    final_layer_type: 'wavenet'
+    target: 'mel' # mel or codec
+    content_dim: 768
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: false
+    n_f0_bins: 512
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: true
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+
+  wavenet:
+    hidden_dim: 768
+    num_layers: 8
+    kernel_size: 5
+    dilation_rate: 1
+    p_dropout: 0.2
+    style_condition: true
+
+loss_params:
+  base_lr: 0.0001

configs/hifigan.yml

@@ -0,0 +1,25 @@
+hift:
+    in_channels: 80
+    base_channels: 512
+    nb_harmonics: 8
+    sampling_rate: 22050
+    nsf_alpha: 0.1
+    nsf_sigma: 0.003
+    nsf_voiced_threshold: 10
+    upsample_rates: [8, 8]
+    upsample_kernel_sizes: [16, 16]
+    istft_params:
+        n_fft: 16
+        hop_len: 4
+    resblock_kernel_sizes: [3, 7, 11]
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]]
+    source_resblock_kernel_sizes: [7, 11]
+    source_resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5]]
+    lrelu_slope: 0.1
+    audio_limit: 0.99
+f0_predictor:
+    num_class: 1
+    in_channels: 80
+    cond_channels: 512
+
+pretrained_model_path: "checkpoints/hift.pt"

configs/inuse/config.json

@@ -0,0 +1 @@
+{"reference_audio_path": "D:/seed-vc/examples/reference/trump_0.wav", "sg_hostapi": "MME", "sg_wasapi_exclusive": false, "sg_input_device": "\u9ea6\u514b\u98ce (Razer BlackShark V2 HS USB", "sg_output_device": "\u626c\u58f0\u5668 (Razer BlackShark V2 HS USB", "sr_type": "sr_model", "diffusion_steps": 8.0, "inference_cfg_rate": 0.7, "max_prompt_length": 3.0, "block_time": 0.58, "crossfade_length": 0.04, "extra_time_ce": 2.5, "extra_time": 0.5, "extra_time_right": 0.02}

configs/presets/config_dit_mel_seed_uvit_whisper_base_f0_44k.yml

@@ -0,0 +1,98 @@
+log_dir: "./runs"
+save_freq: 1
+log_interval: 10
+save_interval: 1000
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 1
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: "DiT_seed_v2_uvit_whisper_base_f0_44k_bigvgan_pruned_ft_ema.pth"
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+preprocess_params:
+  sr: 44100
+  spect_params:
+    n_fft: 2048
+    win_length: 2048
+    hop_length: 512
+    n_mels: 128
+    fmin: 0
+    fmax: "None"
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l1" # l1 or l2
+
+  timbre_shifter:
+    se_db_path: "./modules/openvoice/checkpoints_v2/converter/se_db.pt"
+    ckpt_path: './modules/openvoice/checkpoints_v2/converter'
+
+  vocoder:
+    type: "bigvgan"
+    name: "nvidia/bigvgan_v2_44khz_128band_512x"
+
+  speech_tokenizer:
+    type: 'whisper'
+    name: "openai/whisper-small"
+
+  style_encoder:
+    dim: 192
+    campplus_path: "campplus_cn_common.bin"
+
+  DAC:
+    encoder_dim: 64
+    encoder_rates: [2, 5, 5, 6]
+    decoder_dim: 1536
+    decoder_rates: [ 6, 5, 5, 2 ]
+    sr: 24000
+
+  length_regulator:
+    channels: 768
+    is_discrete: false
+    in_channels: 768
+    content_codebook_size: 2048
+    sampling_ratios: [1, 1, 1, 1]
+    vector_quantize: false
+    n_codebooks: 1
+    quantizer_dropout: 0.0
+    f0_condition: true
+    n_f0_bins: 256
+
+  DiT:
+    hidden_dim: 768
+    num_heads: 12
+    depth: 17
+    class_dropout_prob: 0.1
+    block_size: 8192
+    in_channels: 128
+    style_condition: true
+    final_layer_type: 'mlp'
+    target: 'mel' # mel or codec
+    content_dim: 768
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: true
+    n_f0_bins: 256
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: false
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+    time_as_token: false
+    style_as_token: false
+    uvit_skip_connection: true
+    add_resblock_in_transformer: false
+
+  wavenet:
+    hidden_dim: 768
+    num_layers: 8
+    kernel_size: 5
+    dilation_rate: 1
+    p_dropout: 0.2
+    style_condition: true
+
+loss_params:
+  base_lr: 0.0001
+  lambda_mel: 45
+  lambda_kl: 1.0

configs/presets/config_dit_mel_seed_uvit_whisper_small_wavenet.yml

@@ -0,0 +1,91 @@
+log_dir: "./runs"
+save_freq: 1
+log_interval: 10
+save_interval: 1000
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 2
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: "DiT_seed_v2_uvit_whisper_small_wavenet_bigvgan_pruned.pth"
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+preprocess_params:
+  sr: 22050
+  spect_params:
+    n_fft: 1024
+    win_length: 1024
+    hop_length: 256
+    n_mels: 80
+    fmin: 0
+    fmax: "None"
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l1" # l1 or l2
+
+  timbre_shifter:
+    se_db_path: "./modules/openvoice/checkpoints_v2/converter/se_db.pt"
+    ckpt_path: './modules/openvoice/checkpoints_v2/converter'
+
+  speech_tokenizer:
+    type: 'whisper'
+    name: "openai/whisper-small"
+
+  style_encoder:
+    dim: 192
+    campplus_path: "campplus_cn_common.bin"
+
+  vocoder:
+    type: "bigvgan"
+    name: "nvidia/bigvgan_v2_22khz_80band_256x"
+
+  length_regulator:
+    channels: 512
+    is_discrete: false
+    in_channels: 768
+    content_codebook_size: 2048
+    sampling_ratios: [1, 1, 1, 1]
+    vector_quantize: false
+    n_codebooks: 1
+    quantizer_dropout: 0.0
+    f0_condition: false
+    n_f0_bins: 512
+
+  DiT:
+    hidden_dim: 512
+    num_heads: 8
+    depth: 13
+    class_dropout_prob: 0.1
+    block_size: 8192
+    in_channels: 80
+    style_condition: true
+    final_layer_type: 'wavenet'
+    target: 'mel' # mel or codec
+    content_dim: 512
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: false
+    n_f0_bins: 512
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: true
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+    time_as_token: false
+    style_as_token: false
+    uvit_skip_connection: true
+    add_resblock_in_transformer: false
+
+  wavenet:
+    hidden_dim: 512
+    num_layers: 8
+    kernel_size: 5
+    dilation_rate: 1
+    p_dropout: 0.2
+    style_condition: true
+
+loss_params:
+  base_lr: 0.0001
+  lambda_mel: 45
+  lambda_kl: 1.0

configs/presets/config_dit_mel_seed_uvit_xlsr_tiny.yml

@@ -0,0 +1,82 @@
+log_dir: "./runs/"
+save_freq: 1
+log_interval: 10
+save_interval: 500
+device: "cuda"
+epochs: 1000 # number of epochs for first stage training (pre-training)
+batch_size: 2
+batch_length: 100 # maximum duration of audio in a batch (in seconds)
+max_len: 80 # maximum number of frames
+pretrained_model: "DiT_uvit_tat_xlsr_ema.pth"
+pretrained_encoder: ""
+load_only_params: False # set to true if do not want to load epoch numbers and optimizer parameters
+
+preprocess_params:
+  sr: 22050
+  spect_params:
+    n_fft: 1024
+    win_length: 1024
+    hop_length: 256
+    n_mels: 80
+    fmin: 0
+    fmax: 8000
+
+model_params:
+  dit_type: "DiT" # uDiT or DiT
+  reg_loss_type: "l1" # l1 or l2
+  diffusion_type: "flow"
+
+  timbre_shifter:
+    se_db_path: "./modules/openvoice/checkpoints_v2/converter/se_db.pt"
+    ckpt_path: './modules/openvoice/checkpoints_v2/converter'
+
+  vocoder:
+    type: "hifigan"
+
+  speech_tokenizer:
+    type: 'xlsr'
+    output_layer: 12
+    name: 'facebook/wav2vec2-xls-r-300m'
+
+  style_encoder:
+    dim: 192
+    campplus_path: "campplus_cn_common.bin"
+
+  length_regulator:
+    channels: 384
+    is_discrete: false
+    in_channels: 1024
+    content_codebook_size: 1024
+    sampling_ratios: [1, 1, 1, 1]
+    vector_quantize: false
+    n_codebooks: 2
+    quantizer_dropout: 0.0
+    f0_condition: false
+    n_f0_bins: 512
+
+  DiT:
+    hidden_dim: 384
+    num_heads: 6
+    depth: 9
+    class_dropout_prob: 0.1
+    block_size: 8192
+    in_channels: 80
+    style_condition: true
+    final_layer_type: 'mlp'
+    target: 'mel' # mel or betavae
+    content_dim: 384
+    content_codebook_size: 1024
+    content_type: 'discrete'
+    f0_condition: false
+    n_f0_bins: 512
+    content_codebooks: 1
+    is_causal: false
+    long_skip_connection: false
+    zero_prompt_speech_token: false # for prompt component, do not input corresponding speech token
+    time_as_token: true
+    style_as_token: true
+    uvit_skip_connection: true
+    add_resblock_in_transformer: false
+
+loss_params:
+  base_lr: 0.0001