Bad inference for short sentences

[cdlIris]

Hi,

I have been encountered this problem, while testing with the pre-trained model, and the model that I trained with my own dataset. When the sentences contain 2 or more words, it will work correctly most of the time. However, when the sentence is 1 word, like "Hello!", "Hi", it will generate unintelligible result.

Here are some plots of sentence "Hello!"

Some Background:

• I used my own dataset to train the model, based on the pre-trained model published from this repo.
• has been trained 101k
• Train loss 14000 0.146386, Grad Norm 0.442730, Validation loss 14000: 0.105691

` ################################

Experiment Parameters

    ################################
    epochs=1000,
    iters_per_checkpoint=1000,
    seed=1234,
    dynamic_loss_scaling=True,
    fp16_run=False,
    distributed_run=False,
    dist_backend="nccl",
    dist_url="tcp://localhost:54321",
    cudnn_enabled=True,
    cudnn_benchmark=False,
    ignore_layers=['embedding.weight'],

    ################################
    # Data Parameters             #
    ################################
    load_mel_from_disk=False,
    training_files='filelists/azure0_audio_text_train_filelist.txt',
    validation_files='filelists/azure0_audio_text_val_filelist.txt',
    text_cleaners=['english_cleaners'],

    ################################
    # Audio Parameters             #
    ################################
    max_wav_value=32768.0,
    sampling_rate=22050,
    filter_length=1024,
    hop_length=256,
    win_length=1024,
    n_mel_channels=80,
    mel_fmin=0.0,
    mel_fmax=8000.0,

    ################################
    # Model Parameters             #
    ################################
    n_symbols=len(symbols),
    symbols_embedding_dim=512,

    # Encoder parameters
    encoder_kernel_size=5,
    encoder_n_convolutions=3,
    encoder_embedding_dim=512,

    # Decoder parameters
    n_frames_per_step=1,  # currently only 1 is supported
    decoder_rnn_dim=1024,
    prenet_dim=256,
    max_decoder_steps=2000,
    gate_threshold=0.5,
    p_attention_dropout=0.1,
    p_decoder_dropout=0.1,

    # Attention parameters
    attention_rnn_dim=1024,
    attention_dim=128,

    # Location Layer parameters
    attention_location_n_filters=32,
    attention_location_kernel_size=31,

    # Mel-post processing network parameters
    postnet_embedding_dim=512,
    postnet_kernel_size=5,
    postnet_n_convolutions=5,

    ################################
    # Optimization Hyperparameters #
    ################################
    use_saved_learning_rate=False,
    learning_rate=1e-3,
    weight_decay=1e-6,
    grad_clip_thresh=1.0,
    batch_size=16,
    mask_padding=True  # set model's padded outputs to padded values`

[Abhisitthqgam]

You may need to adjust max_decoder_steps decreased