配置: conf/tuning/train_fastspeech2.yaml
打印配置: false
日志级别: INFO
空运行: false
迭代器类型: sequence
输出目录: exp/a/tts_train_fastspeech2_raw_phn_none
GPU数量: 1
随机种子: 0
工作线程数: 1
注意力绘图数: 3
分布式后端: nccl
分布式初始化方法: env://
分布式世界大小: null
分布式排名: null
本地排名: 0
分布式主地址: null
分布式主端口: null
分布式启动器: null
多进程分布式: false
未使用参数: false
分片DDP: false
启用CuDNN: true
CuDNN基准测试: false
CuDNN确定性: true
收集统计信息: false
写入收集的特征: false
最大轮次: 100
耐心: null
验证调度器标准:
- valid
- loss
早停标准:
- valid
- loss
- min
最佳模型标准:
-   - valid
    - loss
    - min
-   - train
    - loss
    - min
保留最佳模型数: 5
最佳平均间隔: 0
梯度裁剪: 1.0
梯度裁剪类型: 2.0
梯度噪声: false
梯度累积: 8
不运行前向传播: false
恢复: true
训练数据类型: float32
使用AMP: false
日志间隔: null
使用Matplotlib: true
使用TensorBoard: true
使用WandB: false
WandB项目: null
WandB ID: null
WandB实体: null
WandB名称: null
WandB模型日志间隔: -1
检测异常: false
预训练路径: null
初始化参数: []
忽略初始化不匹配: false
冻结参数: []
每轮迭代次数: 800
批量大小: 20
验证批量大小: null
批量箱数: 2500000
验证批量箱数: null
训练形状文件:
- exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/text_shape.phn
- exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/speech_shape
验证形状文件:
- exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/text_shape.phn
- exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/speech_shape
批量类型: numel
验证批量类型: null
折叠长度:
- 150
- 204800
批量内排序: descending
批量排序: descending
多迭代器: false
块长度: 500
块移动比例: 0.5
缓存块数: 1024
训练数据路径和名称及类型:
-   - dump/raw/train_a_phn/text
    - text
    - text
-   - exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/train_a_phn/durations
    - durations
    - text_int
-   - dump/raw/train_a_phn/wav.scp
    - speech
    - sound
验证数据路径和名称及类型:
-   - dump/raw/dev_a_phn/text
    - text
    - text
-   - exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/dev_a_phn/durations
    - durations
    - text_int
-   - dump/raw/dev_a_phn/wav.scp
    - speech
    - sound
允许可变数据键: false
最大缓存大小: 0.0
最大缓存文件描述符: 32
验证最大缓存大小: null
优化器: adam
优化器配置:
    lr: 1.0
调度器: noamlr
调度器配置:
    model_size: 384
    warmup_steps: 4000
令牌列表:
- <blank>
- <unk>
- ','
- .
- r
- t
- n
- a0
- s
- I0
- D
- l
- m
- Y0
- v
- h
- E1
- k
- a:1
- E:1
- G
- f
- j
- T
- a1
- p
- c
- au:1
- i:1
- O:1
- I:1
- E0
- I1
- r_0
- t_h
- k_h
- Y1
- ei1
- i0
- ou:1
- ei:1
- u:1
- O1
- N
- l_0
- '91'
- ai0
- au1
- ou0
- n_0
- ei0
- ai:1
- O0
- ou1
- ai1
- i1
- '9:1'
- '90'
- au0
- x
- c_h
- 9i:1
- C
- p_h
- u0
- Y:1
- J
- 9i1
- u1
- 9i0
- N_0
- m_0
- J_0
- Yi0
- Oi1
- Yi1
- Oi0
- au:0
- '9:0'
- E:0
- <sos/eos>
输出维度: null
模型配置: {}
使用预处理器: true
令牌类型: phn
BPE模型: null
非语言符号: null
清理器: null
G2P: null
特征提取: fbank
特征提取配置:
    n_fft: 1024
    hop_length: 256
    win_length: null
    fs: 22050
    fmin: 80
    fmax: 7600
    n_mels: 80
归一化: global_mvn
归一化配置:
    stats_file: exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/feats_stats.npz
TTS: fastspeech2
TTS配置:
    adim: 384
    aheads: 2
    elayers: 4
    eunits: 1536
    dlayers: 4
    dunits: 1536
    positionwise_layer_type: conv1d
    positionwise_conv_kernel_size: 3
    duration_predictor_layers: 2
    duration_predictor_chans: 256
    duration_predictor_kernel_size: 3
    postnet_layers: 5
    postnet_filts: 5
    postnet_chans: 256
    use_masking: true
    use_scaled_pos_enc: true
    encoder_normalize_before: true
    decoder_normalize_before: true
    reduction_factor: 1
    init_type: xavier_uniform
    init_enc_alpha: 1.0
    init_dec_alpha: 1.0
    transformer_enc_dropout_rate: 0.2
    transformer_enc_positional_dropout_rate: 0.2
    transformer_enc_attention_dropout_rate: 0.2
    transformer_dec_dropout_rate: 0.2
    transformer_dec_positional_dropout_rate: 0.2
    transformer_dec_attention_dropout_rate: 0.2
    pitch_predictor_layers: 5
    pitch_predictor_chans: 256
    pitch_predictor_kernel_size: 5
    pitch_predictor_dropout: 0.5
    pitch_embed_kernel_size: 1
    pitch_embed_dropout: 0.0
    stop_gradient_from_pitch_predictor: true
    energy_predictor_layers: 2
    energy_predictor_chans: 256
    energy_predictor_kernel_size: 3
    energy_predictor_dropout: 0.5
    energy_embed_kernel_size: 1
    energy_embed_dropout: 0.0
    stop_gradient_from_energy_predictor: false
音高提取: dio
音高提取配置:
    fs: 22050
    n_fft: 1024
    hop_length: 256
    f0max: 400
    f0min: 80
    reduction_factor: 1
音高归一化: global_mvn
音高归一化配置:
    stats_file: exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/pitch_stats.npz
能量提取: energy
能量提取配置:
    fs: 22050
    n_fft: 1024
    hop_length: 256
    win_length: null
    reduction_factor: 1
能量归一化: global_mvn
能量归一化配置:
    stats_file: exp/a/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/energy_stats.npz
必需:
- output_dir
- token_list
版本: 0.10.7a1
分布式: false