jik876/hifi-gan issues and pull requests

#100 - Discrimator Loss Too Small

Issue - State: open - Opened by MMingabc almost 3 years ago - 4 comments

#99 - can't synthesize using tacotron2

Issue - State: closed - Opened by cdlIris about 3 years ago - 1 comment

#98 - How to decide when to stop training?

Issue - State: closed - Opened by JohnHerry about 3 years ago - 2 comments

#97 - About inference speed of the three models

Issue - State: open - Opened by JohnHerry about 3 years ago

#96 - How to draw the "Pixel-wise difference in the mel-spectrogram domain " picture?

Issue - State: closed - Opened by JohnHerry about 3 years ago - 5 comments

#95 - Training multi-speaker Hifi-GAN

Issue - State: closed - Opened by JohnHerry about 3 years ago - 3 comments

#94 - Universal_V3?

Issue - State: open - Opened by tylerweitzman about 3 years ago - 1 comment

#93 - The baseline model WaveNet(MoL) is the WaveNet or Parallel WaveNet?

Issue - State: open - Opened by caixxiong about 3 years ago

#92 - apex training?

Issue - State: open - Opened by rave974 about 3 years ago

#91 - 48k config

Issue - State: open - Opened by zpcoftts about 3 years ago

#90 - Pad audio fragment

Issue - State: open - Opened by Alexey322 about 3 years ago - 2 comments

#89 - Error when read wav files by using scipy.io.wavfile

Issue - State: open - Opened by hoang98long about 3 years ago - 1 comment

#88 - How to train and use trained vocoder with fastspeech2

Issue - State: open - Opened by rspiewak47 about 3 years ago - 1 comment

#87 - GPU memory allocation

Issue - State: closed - Opened by rspiewak47 about 3 years ago - 7 comments

#86 - receptive field of the generator

Issue - State: closed - Opened by Liujingxiu23 about 3 years ago - 1 comment

#85 - difficulty of training Universal model with fmax is null

Issue - State: open - Opened by YutoNishimura-v2 about 3 years ago - 8 comments

#84 - How much data is enough?

Issue - State: open - Opened by Mikefizzy over 3 years ago - 5 comments

#83 - Data for MOS computation

Issue - State: open - Opened by AndreevP over 3 years ago

#82 - What is the proper weight for feature-matching loss?

Issue - State: open - Opened by faradayin over 3 years ago - 1 comment

#81 - Inconsistency between the model parameters in the paper and the implementation on the github

Issue - State: open - Opened by Alexey322 over 3 years ago - 2 comments

#80 - the fmax

Issue - State: open - Opened by burcehan over 3 years ago - 2 comments

#79 - What the batch size was used in paper?

Issue - State: open - Opened by Oktai15 over 3 years ago - 7 comments

#78 - Output of the discriminators.

Issue - State: closed - Opened by helloybz over 3 years ago - 1 comment

#77 - Why set the train_loader's shuffle is False ?

Issue - State: closed - Opened by yingfenging over 3 years ago - 2 comments

#76 - Clarification on Finetuning and "Teacher-Forcing"

Issue - State: open - Opened by TaoTeCha over 3 years ago - 1 comment

#75 - Fixed ModuleNotFoundError: No module named 'numba.decorators'

Pull Request - State: open - Opened by SamuelLarkin over 3 years ago - 1 comment

#74 - Can you describe how you achieve so blazing fast inference on CPU?

Issue - State: open - Opened by creotiv over 3 years ago - 1 comment

#73 - What is num_freq?

Issue - State: closed - Opened by wizardk over 3 years ago - 2 comments

#72 - Distributed training

Issue - State: open - Opened by 980202006 over 3 years ago - 3 comments

#71 - Finetuning process

Issue - State: closed - Opened by francislata over 3 years ago - 1 comment

#70 - Fine-tuning from pretrained models?

Issue - State: open - Opened by turian over 3 years ago - 14 comments

#69 - The effect of audio Split or not split?

Issue - State: open - Opened by WhiteFu over 3 years ago - 1 comment

#68 - Quality for some speakers gets worse and worse after 300k steps

Issue - State: open - Opened by ghost over 3 years ago - 1 comment

#67 - What is the Universal model?

Issue - State: closed - Opened by turian over 3 years ago - 4 comments

#66 - Recommended parameters for 48Khz singing synthesis?

Issue - State: closed - Opened by jmasterx over 3 years ago - 1 comment

#65 - Question about training time

Issue - State: open - Opened by max-padgett over 3 years ago - 3 comments

#64 - Questions about MPD's ability to enhance MOS and contribution of feature-matching loss

Issue - State: closed - Opened by francislata over 3 years ago - 1 comment

#63 - Do we need pad audio when train hifigan on tacotron outputs?

Issue - State: closed - Opened by Alexey322 over 3 years ago - 3 comments

#62 - GPU Requirements

Issue - State: closed - Opened by patdflynn over 3 years ago - 1 comment

#61 - Mel spectograms from TensorflowTTS/tacotron2

Issue - State: open - Opened by riverphoenix over 3 years ago - 1 comment

#60 - Validation loss decreasing but training loss is fluctuating a lot.

Issue - State: open - Opened by kannadaraj over 3 years ago - 1 comment

#59 - Training loss plots

Issue - State: open - Opened by ohadvb over 3 years ago - 1 comment

#58 - What are the correct model changes for sample rate 44100?

Issue - State: open - Opened by Alexey322 over 3 years ago - 1 comment

#57 - Any samples with universal pre-trained model?

Issue - State: open - Opened by jun-danieloh over 3 years ago - 1 comment

#56 - error when finetune hifigan with mel generated by fastspeech

Issue - State: closed - Opened by yt605155624 over 3 years ago - 1 comment

#55 - How to train on 48khz data?

Issue - State: closed - Opened by 980202006 over 3 years ago - 3 comments

#54 - Minimum hours of data required for fine-tuning for a single unseen speaker

Issue - State: open - Opened by balag59 over 3 years ago - 9 comments

#53 - Padding mismatch from Tacotron2 pre-processing

Issue - State: closed - Opened by SiddGururani over 3 years ago - 4 comments

#52 - Can't train with output of Tacotron 2

Issue - State: open - Opened by tuong-olli over 3 years ago - 7 comments

#51 - Tracing to torchscript

Issue - State: open - Opened by ctlaltdefeat over 3 years ago - 14 comments

#50 - Batch synthesis noise at the end

Issue - State: closed - Opened by ctlaltdefeat over 3 years ago - 3 comments

#49 - RuntimeError when finetuning the model

Issue - State: closed - Opened by Charlottecuc over 3 years ago - 2 comments

#48 - slope for last activation is different from the rest ones

Issue - State: open - Opened by nukes over 3 years ago - 2 comments

#47 - avoid hard-coding for generator input mel dimensions

Pull Request - State: open - Opened by nukes over 3 years ago

#46 - Will open source data UNIVERSAL_V1？

Issue - State: closed - Opened by 980202006 over 3 years ago - 2 comments

#45 - HiFi-GAN TFLite Model

Issue - State: closed - Opened by tulasiram58827 over 3 years ago - 2 comments

#44 - RuntimeError: Input type (torch.cuda.HalfTensor) and weight type (torch.cuda.FloatTensor) should be the same

Issue - State: closed - Opened by leminhnguyen over 3 years ago - 13 comments

#43 - question about the finetune data

Issue - State: closed - Opened by yanyanxixi over 3 years ago - 10 comments

#42 - machine noise on 4k frequency

Issue - State: open - Opened by hdmjdp over 3 years ago - 8 comments

#41 - Buzzing sound when using Tacotron2+HiFi-GAN

Issue - State: closed - Opened by tulasiram58827 over 3 years ago - 14 comments

#40 - Can the MEL spectrum be resampled?

Issue - State: closed - Opened by 980202006 over 3 years ago - 2 comments

#39 - ExponentialLR and fine tune

Issue - State: open - Opened by Alexey322 over 3 years ago - 3 comments

#38 - how to set segment_size in different sampling_rate

Issue - State: open - Opened by Mingrg over 3 years ago - 2 comments

#37 - How to train a new model with 8k data

Issue - State: closed - Opened by lightwithshadow over 3 years ago - 2 comments

#36 - Some questions about training

Issue - State: closed - Opened by Alexey322 over 3 years ago - 2 comments

#35 - MPD Vs Filter-bank discriminator

Issue - State: open - Opened by ahmed-fau over 3 years ago - 1 comment

#34 - when fine_tuning with the tacotron2 model to produce mel_**.npy, a RuntimeError happened.

Issue - State: closed - Opened by yBeOne over 3 years ago - 2 comments

#33 - Output audio duration does not exactly match input audio.

Issue - State: open - Opened by shawnbzhang almost 4 years ago - 9 comments

#32 - What is the input format?

Issue - State: closed - Opened by prattcmp almost 4 years ago - 4 comments

#31 - mel spectrum loss VS stft loss

Issue - State: closed - Opened by nukes almost 4 years ago - 2 comments

#30 - FineTuning HiFi with GLowTTS npy

Issue - State: open - Opened by 2Bye almost 4 years ago - 8 comments

#29 - Why use those mel spectrogram configurations?

Issue - State: closed - Opened by prattcmp almost 4 years ago - 2 comments

#28 - Specific environment version

Issue - State: closed - Opened by peacefulxy almost 4 years ago - 4 comments

#27 - Why does the mel.npy file I trained with the tacotron2 model not match the dimensions in hifi-gan?

Issue - State: closed - Opened by c9412600 almost 4 years ago - 5 comments

#26 - break point problem

Issue - State: closed - Opened by hdmjdp almost 4 years ago - 14 comments

#25 - Update train.py

Pull Request - State: open - Opened by Miralan almost 4 years ago - 3 comments

#24 - [Question] Dataset preprocessing

Issue - State: closed - Opened by Kreevoz almost 4 years ago - 5 comments

#23 - Just add notes about audio spec.

Pull Request - State: open - Opened by xwaeaewcrhomesysplug almost 4 years ago

#22 - Sorry,a newbie here asking basic setup infos

Issue - State: closed - Opened by xwaeaewcrhomesysplug almost 4 years ago - 5 comments

#21 - Training times v1 vs. v2 vs. v3?

Issue - State: closed - Opened by serg06 almost 4 years ago - 2 comments

#20 - Add the torch pip install full command

Issue - State: closed - Opened by xwaeaewcrhomesysplug almost 4 years ago - 2 comments

#19 - Buzzing noise

Issue - State: closed - Opened by burcehan almost 4 years ago - 3 comments

#18 - log mel spectrum and linear mel spectrum

Issue - State: closed - Opened by burcehan almost 4 years ago - 2 comments

#17 - the loss doesn't decrease

Issue - State: closed - Opened by burcehan almost 4 years ago - 3 comments

#16 - high-frequency has machine noise

Issue - State: closed - Opened by hdmjdp almost 4 years ago - 6 comments

#15 - Glow-tts + hifi-gan inference issue

Issue - State: closed - Opened by Laay almost 4 years ago - 3 comments

#14 - pretrained model

Issue - State: closed - Opened by burcehan almost 4 years ago - 4 comments

#13 - The training time

Issue - State: closed - Opened by burcehan almost 4 years ago - 2 comments

#12 - [Question] How to apply on 16k data?

Issue - State: closed - Opened by OnceJune almost 4 years ago - 15 comments

#11 - [Question] Quality improvement using different generator parameters

Issue - State: closed - Opened by george-roussos almost 4 years ago - 17 comments

#10 - error when fine-tuning

Issue - State: closed - Opened by JoeyHeisenberg almost 4 years ago - 10 comments

#9 - Logging validation loss to tensorboard

Issue - State: closed - Opened by rafaelvalle almost 4 years ago - 1 comment

#8 - window size is hardcoded (44khz HiFi-GAN)

Issue - State: closed - Opened by rafaelvalle almost 4 years ago - 2 comments

#7 - v1 Training time + training a 44Khz model on Universal VCTK+Blizzard2011+Clipper Datasets

Issue - State: closed - Opened by CookiePPP almost 4 years ago - 18 comments

#6 - add support for 2D Mel Spec during Fine tuning

Pull Request - State: closed - Opened by Edresson almost 4 years ago - 1 comment

#5 - Discriminator weights for pretrained models

Issue - State: closed - Opened by ctlaltdefeat almost 4 years ago - 2 comments

#4 - Some questions

Issue - State: open - Opened by george-roussos almost 4 years ago - 41 comments

GitHub / jik876/hifi-gan issues and pull requests