From 2b529f60c84e9da1b3064a3799716aff4ad6d5f4 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Eren=20G=C3=B6lge?= Date: Fri, 9 Apr 2021 11:40:06 +0200 Subject: [PATCH] update default hifigan config --- ...ied_hifigan.json => hifigan_ljspeech.json} | 49 ++++++++++--------- 1 file changed, 27 insertions(+), 22 deletions(-) rename TTS/vocoder/configs/{modified_hifigan.json => hifigan_ljspeech.json} (77%) diff --git a/TTS/vocoder/configs/modified_hifigan.json b/TTS/vocoder/configs/hifigan_ljspeech.json similarity index 77% rename from TTS/vocoder/configs/modified_hifigan.json rename to TTS/vocoder/configs/hifigan_ljspeech.json index 5851d2b4..23cbf3f8 100644 --- a/TTS/vocoder/configs/modified_hifigan.json +++ b/TTS/vocoder/configs/hifigan_ljspeech.json @@ -11,10 +11,10 @@ "frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used. // Audio processing parameters - "sample_rate": 16000, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled. + "sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled. "preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis. "ref_level_db": 20, // reference level db, theoretically 20db is the sound of air. - "log_func": "np.log", + "log_func": "np.log10", "do_sound_norm": true, // Silence trimming @@ -23,17 +23,17 @@ // MelSpectrogram parameters "num_mels": 80, // size of the mel spec frame. - "mel_fmin": 0.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!! - "mel_fmax": 8000.0, // maximum freq level for mel-spec. Tune for dataset!! + "mel_fmin": 50.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!! + "mel_fmax": 7600.0, // maximum freq level for mel-spec. Tune for dataset!! "spec_gain": 1.0, // scaler value appplied after log transform of spectrogram. // Normalization parameters - "signal_norm": false, // normalize spec values. Mean-Var normalization if 'stats_path' is defined otherwise range normalization defined by the other params. + "signal_norm": true, // normalize spec values. Mean-Var normalization if 'stats_path' is defined otherwise range normalization defined by the other params. "min_level_db": -100, // lower bound for normalization "symmetric_norm": true, // move normalization to range [-1, 1] "max_norm": 4.0, // scale normalization to range [-max_norm, max_norm] or [0, max_norm] "clip_norm": true, // clip normalized values into the range. - "stats_path": null // DO NOT USE WITH MULTI_SPEAKER MODEL. scaler stats file computed by 'compute_statistics.py'. If it is defined, mean-std based notmalization is used and other normalization params are ignored + "stats_path": "/home/erogol/.local/share/tts/tts_models--en--ljspeech--speedy-speech-wn/scale_stats.npy" }, // DISTRIBUTED TRAINING @@ -70,7 +70,7 @@ "l1_spec_loss_params": { "use_mel": true, - "sample_rate": 16000, + "sample_rate": 22050, "n_fft": 1024, "hop_length": 256, "win_length": 1024, @@ -104,7 +104,7 @@ }, // DATASET - "data_path": "/home/erogol/gdrive/Datasets/non-binary-voice-files/vo_voice_quality_transformation/", + "data_path": "/home/erogol/gdrive/Datasets/LJSpeech-1.1/wavs/", "feature_path": null, // "feature_path": "/home/erogol/gdrive/Datasets/non-binary-voice-files/tacotron-DCA/", "seq_len": 8192, @@ -127,25 +127,30 @@ "wd": 0.0, // Weight decay weight. "gen_clip_grad": -1, // Generator gradient clipping threshold. Apply gradient clipping if > 0 "disc_clip_grad": -1, // Discriminator gradient clipping threshold. - // "lr_scheduler_gen": "ExponentialLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate - // "lr_scheduler_gen_params": { - // "gamma": 0.999, - // "last_epoch": -1 - // }, - // "lr_scheduler_disc": "ExponentialLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate - // "lr_scheduler_disc_params": { - // "gamma": 0.999, - // "last_epoch": -1 - // }, - "lr_gen": 0.00001, // Initial learning rate. If Noam decay is active, maximum learning rate. - "lr_disc": 0.00001, + "lr_gen": 0.0002, // Initial learning rate. If Noam decay is active, maximum learning rate. + "lr_disc": 0.0002, + "optimizer": "AdamW", + "optimizer_params":{ + "betas": [0.8, 0.99], + "weight_decay": 0.0 + }, + "lr_scheduler_gen": "ExponentialLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate + "lr_scheduler_gen_params": { + "gamma": 0.999, + "last_epoch": -1 + }, + "lr_scheduler_disc": "ExponentialLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate + "lr_scheduler_disc_params": { + "gamma": 0.999, + "last_epoch": -1 + }, // TENSORBOARD and LOGGING "print_step": 25, // Number of steps to log traning on console. "print_eval": false, // If True, it prints loss values for each step in eval run. "save_step": 25000, // Number of training steps expected to plot training stats on TB and save model checkpoints. "checkpoint": true, // If true, it saves checkpoints per "save_step" - "tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging. + "tb_model_param_stats": true, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging. // DATA LOADING "num_loader_workers": 8, // number of training data loader processes. Don't set it too big. 4-8 are good values. @@ -153,7 +158,7 @@ "eval_split_size": 10, // PATHS - "output_path": "/home/erogol/gdrive/Trainings/sam/" + "output_path": "/home/erogol/gdrive/Trainings/LJSpeech/" }