Crashing while saving checkpoint

[astricks]

Hi,

I am trying to train a Tacotron2 model in Hindi. I have my own 25 hour single speaker cleaned dataset. I'm using the following configuration.

{ "model": "Tacotron2", "run_name": "hindi-ddc", "run_description": "tacotron2 with DDC and differential spectral loss.",

// AUDIO PARAMETERS
"audio":{
    // stft parameters
    "fft_size": 1024,         // number of stft frequency levels. Size of the linear spectogram frame.
    "win_length": 1024,      // stft window length in ms.
    "hop_length": 256,       // stft window hop-lengh in ms.
    "frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
    "frame_shift_ms": null,  // stft window hop-lengh in ms. If null, 'hop_length' is used.

    // Audio processing parameters
    "sample_rate": 22050,   // DATASET-RELATED: wav sample-rate.
    "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.

    // Silence trimming
    "do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (true), TWEB (false), Nancy (true)
    "trim_db": 60,          // threshold for timming silence. Set this according to your dataset.

    // Griffin-Lim
    "power": 1.5,           // value to sharpen wav signals after GL algorithm.
    "griffin_lim_iters": 60,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.

    // MelSpectrogram parameters
    "num_mels": 80,         // size of the mel spec frame.
    "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,

    // Normalization parameters
    "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
},

// VOCABULARY PARAMETERS
// if custom character set is not defined,
// default set in symbols.py is used
"characters":{
    "pad": "_",
    "eos": "~",
    "bos": "^",
    "characters": "अआइईउऊऋएऐऑओऔकखगघङचछजझञटठडढणतथदधनपफबभमयरलवशषसहह़ा",
    "punctuations":"!'\",.:?। ",
    "phonemes":"iyɨʉɯuɪʏʊeøɘəɵɤoɛœɜɞʌɔæɐaɶɑɒᵻʘɓǀɗǃʄǂɠǁʛpbtdʈɖcɟkɡqɢʔɴŋɲɳnɱmʙrʀⱱɾɽɸβfvθðszʃʒʂʐçʝxɣχʁħʕhɦɬɮʋɹɻjɰlɭʎʟˈˌːˑʍwɥʜʢʡɕʑɺɧɚ˞ɫ"
},

// DISTRIBUTED TRAINING
"distributed":{
    "backend": "nccl",
    "url": "tcp:\/\/localhost:54321"
},

"reinit_layers": [],    // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.

// TRAINING
"batch_size": 32,       // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
"eval_batch_size":16,
"r": 7,                 // Number of decoder frames to predict per iteration. Set the initial values if gradual training is enabled.
"gradual_training": [[0, 7, 64], [1, 5, 64], [50000, 3, 32], [130000, 2, 32], [290000, 1, 32]], //set gradual training steps [first_step, r, batch_size]. If it is null, gradual training is disabled. For Tacotron, you might need to reduce the 'batch_size' as you proceeed.
"mixed_precision": true,     // level of optimization with NVIDIA's apex feature for automatic mixed FP16/FP32 precision (AMP), NOTE: currently only O1 is supported, and use "O1" to activate.

// LOSS SETTINGS
"loss_masking": true,       // enable / disable loss masking against the sequence padding.
"decoder_loss_alpha": 0.5,  // original decoder loss weight. If > 0, it is enabled
"postnet_loss_alpha": 0.25, // original postnet loss weight. If > 0, it is enabled
"postnet_diff_spec_alpha": 0.25,     // differential spectral loss weight. If > 0, it is enabled
"decoder_diff_spec_alpha": 0.25,     // differential spectral loss weight. If > 0, it is enabled
"decoder_ssim_alpha": 0.5,     // decoder ssim loss weight. If > 0, it is enabled
"postnet_ssim_alpha": 0.25,     // postnet ssim loss weight. If > 0, it is enabled
"ga_alpha": 5.0,           // weight for guided attention loss. If > 0, guided attention is enabled.
"stopnet_pos_weight": 15.0, // pos class weight for stopnet loss since there are way more negative samples than positive samples.

// VALIDATION
"run_eval": true,
"test_delay_epochs": 10,  //Until attention is aligned, testing only wastes computation time.
"test_sentences_file": null,  // set a file to load sentences to be used for testing. If it is null then we use default english sentences.

// OPTIMIZER
"noam_schedule": false,        // use noam warmup and lr schedule.
"grad_clip": 1.0,              // upper limit for gradients for clipping.
"epochs": 1000,                // total number of epochs to train.
"lr": 0.0001,                  // Initial learning rate. If Noam decay is active, maximum learning rate.
"wd": 0.000001,                // Weight decay weight.
"warmup_steps": 4000,          // Noam decay steps to increase the learning rate from 0 to "lr"
"seq_len_norm": false,         // Normalize eash sample loss with its length to alleviate imbalanced datasets. Use it if your dataset is small or has skewed distribution of sequence lengths.

// TACOTRON PRENET
"memory_size": -1,             // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame.
"prenet_type": "original",     // "original" or "bn".
"prenet_dropout": false,       // enable/disable dropout at prenet.

// TACOTRON ATTENTION
"attention_type": "original",  // 'original' , 'graves', 'dynamic_convolution'
"attention_heads": 4,          // number of attention heads (only for 'graves')
"attention_norm": "sigmoid",   // softmax or sigmoid.
"windowing": false,            // Enables attention windowing. Used only in eval mode.
"use_forward_attn": false,     // if it uses forward attention. In general, it aligns faster.
"forward_attn_mask": false,    // Additional masking forcing monotonicity only in eval mode.
"transition_agent": false,     // enable/disable transition agent of forward attention.
"location_attn": true,         // enable_disable location sensitive attention. It is enabled for TACOTRON by default.
"bidirectional_decoder": false,  // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset.
"double_decoder_consistency": true,  // use DDC explained here https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency-draft/
"ddc_r": 7,                           // reduction rate for coarse decoder.

// STOPNET
"stopnet": true,               // Train stopnet predicting the end of synthesis.
"separate_stopnet": true,      // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.

// TENSORBOARD and LOGGING
"print_step": 25,       // Number of steps to log training on console.
"tb_plot_step": 100,    // Number of steps to plot TB training figures.
"print_eval": false,     // If True, it prints intermediate loss values in evalulation.
"save_step": 200,      // Number of training steps expected to save traninpg stats and checkpoints.
"checkpoint": true,     // If true, it saves checkpoints per "save_step"
"keep_all_best": false,  // If true, keeps all best_models after keep_after steps
"keep_after": 10000,    // Global step after which to keep best models if keep_all_best is true
"tb_model_param_stats": false,     // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.

// DATA LOADING
"text_cleaner": "basic_cleaners",
"enable_eos_bos_chars": false, // enable/disable beginning of sentence and end of sentence chars.
"num_loader_workers": 2,        // number of training data loader processes. Don't set it too big. 4-8 are good values.
"num_val_loader_workers": 2,    // number of evaluation data loader processes.
"batch_group_size": 4,  //Number of batches to shuffle after bucketing.
"min_seq_len": 81,       // DATASET-RELATED: minimum text length to use in training
"max_seq_len": 186,     // DATASET-RELATED: maximum text length
"compute_input_seq_cache": false,  // if true, text sequences are computed before starting training. If phonemes are enabled, they are also computed at this stage.
"use_noise_augment": true,

// PATHS
"output_path": "/home/ubuntu/output/",

// PHONEMES
"phoneme_cache_path": "/home/ubuntu/phoneme_cache/",  // phoneme computation is slow, therefore, it caches results in the given folder.
"use_phonemes": false,           // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "hi",     // depending on your target language, pick one from  https://github.com/bootphon/phonemizer#languages

// MULTI-SPEAKER and GST
"use_speaker_embedding": false,      // use speaker embedding to enable multi-speaker learning.
"use_gst": false,                       // use global style tokens
"use_external_speaker_embedding_file": false, // if true, forces the model to use external embedding per sample instead of nn.embeddings, that is, it supports external embeddings such as those used at: https://arxiv.org/abs /1806.04558
"external_speaker_embedding_file": "../../speakers-vctk-en.json", // if not null and use_external_speaker_embedding_file is true, it is used to load a specific embedding file and thus uses these embeddings instead of nn.embeddings, that is, it supports external embeddings such as those used at: https://arxiv.org/abs /1806.04558
"gst":  {                           // gst parameter if gst is enabled
    "gst_style_input": null,        // Condition the style input either on a
                                    // -> wave file [path to wave] or
                                    // -> dictionary using the style tokens {'token1': 'value', 'token2': 'value'} example {"0": 0.15, "1": 0.15, "5": -0.15}
                                    // with the dictionary being len(dict) <= len(gst_style_tokens).
    "gst_embedding_dim": 512,
    "gst_num_heads": 4,
    "gst_style_tokens": 10,
    "gst_use_speaker_embedding": false
},

// DATASETS
"datasets":   // List of datasets. They all merged and they get different speaker_ids.
    [
        {
            "name": "hindi",
            "path": "/dev/data/hindidataset/",
            "meta_file_train": "metadata.csv", // for vtck if list, ignore speakers id in list for train, its useful for test cloning with new speakers
            "meta_file_val": null
        }
    ]

}

--

The stacktrace I'm hitting is below.

CHECKPOINT : /home/ubuntu/output/modi-ddc-March-19-2021_05+17PM-8545a69/checkpoint_200.pth.tar /home/ubuntu/TTS/TTS/utils/audio.py:234: RuntimeWarning: overflow encountered in power return np.power(10.0, x / self.spec_gain) ! Run is kept in /home/ubuntu/output/modi-ddc-March-19-2021_05+17PM-8545a69 Traceback (most recent call last): File "TTS/bin/train_tacotron.py", line 664, in main(args) File "TTS/bin/train_tacotron.py", line 634, in main scaler_st) File "TTS/bin/train_tacotron.py", line 312, in train train_audio = ap.inv_melspectrogram(const_spec.T) File "/home/ubuntu/TTS/TTS/utils/audio.py", line 286, in inv_melspectrogram return self._griffin_lim(S*self.power) File "/home/ubuntu/TTS/TTS/utils/audio.py", line 315, in _griffin_lim angles = np.exp(1j np.angle(self._stft(y))) File "/home/ubuntu/TTS/TTS/utils/audio.py", line 303, in _stft pad_mode=self.stft_pad_mode, File "/home/ubuntu/.local/lib/python3.6/site-packages/librosa/core/spectrum.py", line 215, in stft util.valid_audio(y) File "/home/ubuntu/.local/lib/python3.6/site-packages/librosa/util/utils.py", line 275, in valid_audio raise ParameterError('Audio buffer is not finite everywhere') librosa.util.exceptions.ParameterError: Audio buffer is not finite everywhere

--

I've been trying to debug for 2 days but not able to make progress. I'd really appreciate any help/suggestions.

[erogol]

Do you have tensorboard outputs for spectrograms?

[astricks]

Please find the tensorflow events file at the link below. Please let me know if I can provide any more information.

https://drive.google.com/drive/folders/1nlpz6uPVaSLqfP61zWF_Po1XFuex4C5z?usp=sharing

[lexkoro]

@astricks Hey I've had a look at your logs and the spectrograms seem broken. Are you sure that your dataset is okay?

One suggestion: You could try setting "spec_gain": 20. With "spec_gain": 1 and not having a stats_file (see stats_path), people had similiar problems regarding the spectrograms in the past.

One question: Is there a reason you set "min_seq_len": 81? Seems a bit high, but I'm not sure in which way that might affect the model.

[astricks]

@SanjaESC I came up with min_seq_len=81 by passing sorting the sentences by length and using awk to calculate the length. Below is the shortest sentence. Passing it through phenomize gives length 82, so figured i'd go with the shorter one. Do these have to be exact numbers?

देशवासियों आप सब को नमस्कार 2015।

I'll try increasing spec_gain to 20 and running again.

[lexkoro]

@SanjaESC I came up with min_seq_len=81 by passing sorting the sentences by length and using awk to calculate the length. Below is the shortest sentence. Passing it through phenomize gives length 82, so figured i'd go with the shorter one. Do these have to be exact numbers?

They don't have to be exact numbers. Those are variable options where you can decide which data to use. So in your case you use data with text length between min. 82 and max. 186 character length. When you start the training it should say at the beginning how many files where filtered based on your min/max settings.

[astricks]

I fixed the min/max lengths (6 and 280) and set spec_gain to 20, and its working fine! Thanks for all the help. I'll share the Hindi model once it's trained. Closing the issue.