This model was trained by kan-bayashi using csmsc/tts1 recipe in espnet.
\n\n\n\n
See https://github.com/espnet/espnet_model_zoogit clone https://github.com/espnet/espnet\ncd espnet\ngit checkout 51352aee9ae318640e128a645e722d1f7524edb1\npip install -e .\ncd egs2/csmsc/tts1\n# Download the model file here\n./run.sh --skip_data_prep false --skip_train true --download_model kan-bayashi/csmsc_tts_train_fastspeech_raw_phn_pypinyin_g2p_phone_train.loss.ave\nconfig: conf/tuning/train_fastspeech.yaml\nprint_config: false\nlog_level: INFO\ndry_run: false\niterator_type: sequence\noutput_dir: exp/tts_train_fastspeech_raw_phn_pypinyin_g2p_phone\nngpu: 1\nseed: 0\nnum_workers: 1\nnum_att_plot: 3\ndist_backend: nccl\ndist_init_method: env://\ndist_world_size: 4\ndist_rank: 0\nlocal_rank: 0\ndist_master_addr: localhost\ndist_master_port: 49841\ndist_launcher: null\nmultiprocessing_distributed: true\ncudnn_enabled: true\ncudnn_benchmark: false\ncudnn_deterministic: true\ncollect_stats: false\nwrite_collected_feats: false\nmax_epoch: 200\npatience: null\nval_scheduler_criterion:\n- valid\n- loss\nearly_stopping_criterion:\n- valid\n- loss\n- min\nbest_model_criterion:\n- - valid\n - loss\n - min\n- - train\n - loss\n - min\nkeep_nbest_models: 5\ngrad_clip: 1.0\ngrad_clip_type: 2.0\ngrad_noise: false\naccum_grad: 1\nno_forward_run: false\nresume: true\ntrain_dtype: float32\nuse_amp: false\nlog_interval: null\npretrain_path: []\npretrain_key: []\nnum_iters_per_epoch: 500\nbatch_size: 20\nvalid_batch_size: null\nbatch_bins: 4800000\nvalid_batch_bins: null\ntrain_shape_file:\n- exp/tts_stats_raw_phn_pypinyin_g2p_phone/train/text_shape.phn\n- exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/tr_no_dev/speech_shape\nvalid_shape_file:\n- exp/tts_stats_raw_phn_pypinyin_g2p_phone/valid/text_shape.phn\n- exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/dev/speech_shape\nbatch_type: numel\nvalid_batch_type: null\nfold_length:\n- 150\n- 800\nsort_in_batch: descending\nsort_batch: descending\nmultiple_iterator: false\nchunk_length: 500\nchunk_shift_ratio: 0.5\nnum_cache_chunks: 1024\ntrain_data_path_and_name_and_type:\n- - dump/raw/tr_no_dev/text\n - text\n - text\n- - exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/tr_no_dev/durations\n - durations\n - text_int\n- - exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/tr_no_dev/denorm/feats.scp\n - speech\n - npy\nvalid_data_path_and_name_and_type:\n- - dump/raw/dev/text\n - text\n - text\n- - exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/dev/durations\n - durations\n - text_int\n- - exp/tts_train_tacotron2_raw_phn_pypinyin_g2p_phone/decode_train.loss.ave/dev/denorm/feats.scp\n - speech\n - npy\nallow_variable_data_keys: false\nmax_cache_size: 0.0\nvalid_max_cache_size: null\noptim: adam\noptim_conf:\n lr: 1.0\nscheduler: noamlr\nscheduler_conf:\n model_size: 384\n warmup_steps: 4000\ntoken_list:\n- \n- \n- d\n- sh\n- j\n- l\n- \"\\u3002\"\n- zh\n- \"\\uFF0C\"\n- i4\n- x\n- h\n- b\n- e\n- g\n- t\n- m\n- z\n- q\n- i1\n- i3\n- ch\n- u4\n- n\n- f\n- i2\n- r\n- k\n- s\n- e4\n- ai4\n- a1\n- c\n- p\n- ian4\n- uo3\n- ao3\n- ai2\n- ao4\n- an4\n- ong1\n- u3\n- ing2\n- en2\n- e2\n- u2\n- ui4\n- ian2\n- iou3\n- ang4\n- u1\n- iao4\n- uo4\n- eng2\n- a4\n- in1\n- eng1\n- ou3\n- ang1\n- ian1\n- ou4\n- ing1\n- uo1\n- an1\n- ian3\n- ie3\n- a3\n- ing4\n- an3\n- an2\n- \"\\xFC4\"\n- iao3\n- ei4\n- ong2\n- en1\n- uei4\n- \"\\xFCan2\"\n- ang2\n- ang3\n- iu4\n- iang4\n- ai3\n- ao1\n- ou1\n- eng4\n- iang3\n- en3\n- ai1\n- ong4\n- ie4\n- e3\n- ia1\n- uo2\n- ia4\n- \"\\xFC3\"\n- uan1\n- er2\n- ei3\n- ei2\n- iang1\n- i\n- ing3\n- en4\n- \"\\xFC2\"\n- uan3\n- e1\n- in2\n- iao1\n- in4\n- ie1\n- ong3\n- iang2\n- ie2\n- uan4\n- a2\n- ui3\n- eng3\n- uan2\n- \"\\xFCe4\"\n- uai4\n- ou2\n- \"\\uFF1F\"\n- \"\\xFCe2\"\n- in3\n- uang3\n- uang1\n- iu2\n- en\n- a\n- ao2\n- ua4\n- un1\n- ui1\n- uei2\n- iong4\n- uang2\n- v3\n- ui2\n- iao2\n- uang4\n- \"\\xFC1\"\n- ei1\n- o2\n- er4\n- iou2\n- iou4\n- \"\\uFF01\"\n- ua1\n- \"\\xFCan4\"\n- iu3\n- un4\n- \"\\xFCan3\"\n- uen2\n- \"\\xFCn4\"\n- iu1\n- un3\n- uen4\n- er3\n- \"\\xFCn1\"\n- un2\n- \"\\xFCn2\"\n- o4\n- o1\n- ua2\n- uei1\n- uei3\n- ia3\n- iong3\n- ua3\n- ia\n- \"\\xFCe1\"\n- v4\n- \"\\xFCan1\"\n- iong1\n- ia2\n- uai1\n- iong2\n- iou1\n- uai3\n- \"\\xFCe3\"\n- uen1\n- uen3\n- uai2\n- o3\n- er\n- ve4\n- io1\n- \"\\xFCn3\"\n- u\n- ou\n- o\n- ang\n- ueng1\n- v2\n- uo\n- ao\n- ueng4\n- ua\n- ei\n- uen\n- an\n- '2'\n- ueng3\n- iang\n- \"\\xFC\"\n- ie\n- \"\\uFF30\"\n- \"\\uFF22\"\n- ai\n- \nodim: 80\nmodel_conf: {}\nuse_preprocessor: true\ntoken_type: phn\nbpemodel: null\nnon_linguistic_symbols: null\ncleaner: null\ng2p: pypinyin_g2p_phone\nfeats_extract: null\nfeats_extract_conf: null\nnormalize: global_mvn\nnormalize_conf:\n stats_file: exp/tts_stats_raw_phn_pypinyin_g2p_phone/train/feats_stats.npz\ntts: fastspeech\ntts_conf:\n adim: 384\n aheads: 2\n elayers: 6\n eunits: 1536\n dlayers: 6\n dunits: 1536\n positionwise_layer_type: conv1d\n positionwise_conv_kernel_size: 3\n duration_predictor_layers: 2\n duration_predictor_chans: 384\n duration_predictor_kernel_size: 3\n postnet_layers: 5\n postnet_filts: 5\n postnet_chans: 256\n use_masking: true\n use_scaled_pos_enc: true\n encoder_normalize_before: true\n decoder_normalize_before: true\n reduction_factor: 1\n init_type: xavier_uniform\n init_enc_alpha: 1.0\n init_dec_alpha: 1.0\n transformer_enc_dropout_rate: 0.1\n transformer_enc_positional_dropout_rate: 0.1\n transformer_enc_attn_dropout_rate: 0.1\n transformer_dec_dropout_rate: 0.1\n transformer_dec_positional_dropout_rate: 0.1\n transformer_dec_attn_dropout_rate: 0.1\npitch_extract: null\npitch_extract_conf: {}\npitch_normalize: null\npitch_normalize_conf: {}\nenergy_extract: null\nenergy_extract_conf: {}\nenergy_normalize: null\nenergy_normalize_conf: {}\nrequired:\n- output_dir\n- token_list\ndistributed: true