TensorFlowTTS

TensorFlowTTS copied to clipboard

Performance Benchmark

1

Is there any data on performance benchmark for various GPU memory / batch size configurations? I am using 15gb single GPU. Unless I use 16 for batch size, it throws...

iprovalo

1. for `HiFiGAN's` stable training, `feature_loss` of `MPD` should be computed, so in the each `PD`, it should return `all convolution's outputs`, not only `the last one`. But in each...

BridgetteSong

Ukrainian TTS

2

Hello. If I will publish a high-quality dataset (around 8 hours of speech) for Ukrainian, then you can make a processor for it?

egorsmkv

Longer Chinese sentences use fastspeech2 to synthesize exceptions

4

Longer Chinese sentences use fastspeech2 to synthesize exceptions. TEXT:在一个拥有 2500 万人口的不发达且以孤立着称的国家，绝大多数人被认为没有接种疫苗，它有可能成为一场规模巨大的人道主义灾难，威胁到几乎任何政府的权力控制 在世界上。 WAVE：[2.zip](https://github.com/TensorSpeech/TensorFlowTTS/files/9138051/2.zip) @dathudeptrai

lawo123

As shown in the figure below, the inference speed of tflite is much slower than that of tensorflow.Why? 

lawo123

Fastspeech2 training error : Not found any mels files in $./dump_ljspeech/train/.

5

Hi all, I have just started learning TTS model building. I am trying to train fastspeech2 model using ljspeech dataset. For that, I have donwloaded ljspeech dataset and run folllowing...

neso613

InvalidArgumentError: Paddings must be non-negative: 0 -13480

1

tensorflow-addons 0.17.1 tensorflow-estimator 2.7.0 tensorflow-gpu 2.7.0 tensorflow-io-gcs-filesystem 0.24.0 ------------ python3 examples/melgan/train_melgan.py \ 2 --train-dir ./dump_baker/train/ \ 3 --dev-dir ./dump_baker/valid/ \ 4 --outdir ./examples/melgan/exp/train.melgan.v2/ \ 5 --config ./examples/melgan/conf/melgan.v1.yaml \ 6 --use-norm...

zhanggang11863976

Hi, I'm training Fastspeech2 with ~20 hours dataset and get tensorboard result as following:  As you can see, f0 and energy is overfitting. What can lead to this problem?

icyda17

Hi! We trained fastspeech2 network with Persian dataset. The point that caught our attention after the training of the network was that if we add english g2p outputs to the...

arefekh

the result is **bad** in tensorflowTTS (fastspeech2) with data `aishell3`， can you updata the code ? 我用aishell3数据训练出来的模型，合成效果感觉挺差的哎。请问能更新一下aishell3的合成吗？

Tian14267