imdanboy

I tried batch size: (32, 128, 256), with similar configuration of this repo, batch 32 was better than others at 220k train step ( 32 > 128 > 256). I...

I just experimentally found batch 16 was best with learning_rate: (discriminator: 4e-4, generator: 1e-4 gan training technique? called ttur was beneficial) if other hyper params are fixed.

If alignment loss doesn't drop in first few epoch, how about scaling an alignment loss term by tuning `lambda_align`? I have experienced an alignment issue in vctk datset so i...

@hermanseu Have you checked an alignment loss or a plotting of alignment matrix? I think abnormal alignment could be one reason for non convergence of mel loss. Whenever diagonal align...

The model is not intended to work on unseen speaker, so it should be trained on target speaker. If you mean 'training from scratch' vs 'fine-tuning' for new speaker, I...

Sorry for late reply. Because I have limited experience, i can say for sure only what i've done. The experiment I've done was conducted on [LJSpeech dataset](https://keithito.com/LJ-Speech-Dataset/) which has 13,100...

Hi, `batch_bins: 3000000` with 4 V100 32GB seems to use almost all memory. Thus how about trying `3000000/4 * (24/32) = 562500` Or reducing the value of `segment_size` which also...

Sure, I think jets also works well for turkce(turkish) language! Although I have trained and shown the result on English and Korean in this repo. In your case, the difference...

Oh, then you should prepare dataset first. (how about searching for publicly available dataset) I don't know whether ESPnet provides turkish tts recipe or not. In espnet, all you have...

Sorry for late, I recently recognized the current implementation regarding on an alignment learning is different from official code [Nvidia-FastPitch](https://github.com/NVIDIA/DeepLearningExamples/blob/9becdf8d53aac528cc5e214e27314f2572971d48/PyTorch/SpeechSynthesis/FastPitch/fastpitch/attention.py#L211) as discussed at https://github.com/espnet/espnet/issues/5179#issuecomment-1565241556 Thanks a lot, I will check...