See this thread
https://groups.google.com/forum/?utm_medium=email&utm_source=footer#!msg/kaldi-help/1AMGQ54tn0E/kacpkoizAAAJ
for the problem-- for CUDA 10.1 the rpath we add for CUDA is not sufficient because they put the CUBLAS library somewhere else. Making the issue so we don't lose track of it.

• 系统环境：

  • Paddle版本：1.5.1，CPU，无使用其他加速模块
  • 系统: CentOS 6.3

• 问题描述：

  • 使用paddle.fluid.contrib.slim.Compressor模块进行模型压缩
  • 压缩后的模型，float能正常运行，int8版出现以下错误：
    

• 问题复现：

git clone https://github.com/Bond-SYSU/paddle_compress.git
cd paddle_compress
sh run.sh compress   #

After training, using some script like this: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/freeze_graph.py ?

What should be the input node name(s) and output node names used for freeze and inference later?

Thanks!

Due to the 2GB contiguous block allocation limitation of 32 bit systems, this line:

https://github.com/readbeyond/aeneas/blob/master/aeneas/audiofile.py#L447

might throw a MemoryError, causing an "unexpected exception".

We should:

1. catch it and log a meaningful error
2. document this limitation somewhere

We would benefit from tutorial how to convert DNN models trained either by HTK 3.5.2-BETA or by Kaldi. I am specifically interested how you converted DNN models trained by Kaldi later used in Japanese speech dictation published here.

🐛 Bug

In order to enable inversion of an STFT via the inverse STFT in istft, it is sufficient that the signal windowing obeys the constraint of “Constant OverLap Add” (COLA). This ensures that every point in the input data is equally weighted, thereby avoiding aliasing and allowing full reconstruction.
https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.check_COLA.html

Le

@lfchener 我看报错这行代码是您最近添加的，我是pip安装的环境， 并按照readme 一步步执行的，但执行到infer时报错，是readme过时了吗？谢谢。
Traceback (most recent call last):
File "/home/shenxz/.IdeaIC2019.2/config/plugins/python-ce/helpers/pydev/pydevd.py", line 1415, in _exec
pydev_imports.execfile(file, globals, locals) # execute the script
File "/home/shenxz/code/DeepSpeech/infer.py", line 152, in
main()
File "/home/sh

It looks like the default initialization scheme with tf.get_variable is uniform Glorot , but the paper uses He's Gaussian.

hi,
as you know, in SoLoud, the number of filters are limited
we should implement more like different reverbs, fir and irr filters, (these could be used to implement HRTF support), Chorus, One Poll, One Zero, Pole Zero, Two Pole, Two Zero, etc
a library exists called stk under zlib license which already implemented these maybe we can implement some of these out

I think the filenames in models.sh referred to on lines 4-9 should refer to kaldi-generic-en-tdnn_f-r20190609* which is downloaded on line 3.

Currently (6.0.15), inline comments (marked by ;) in lines with command names are interpreted to be part of the command name:

Create TextGrid: 0, 1, "tier", "" ; Good
asserterror Command "Remove ; Bad!"
  ... not available for current selection.
Remove ; Bad!

As shown in the snippet, a comment next to the first command (after the argument list) is correctly ignored, but when used

Firebase functions support only 9mins.

https://cloud.google.com/functions/docs/concepts/exec#timeout

Is there any way to use webhook or gcp pubsub for async long job inside gcp functions?

https://cloud.google.com/community/tutorials/cloud-functions-async

Final goal is to produce a functional Cboard app for iOS.
We have to use the Cordova repository called ccboard which is already producing the Android app:
https://github.com/cboard-org/ccboard

• Update readme file to include app compilation and production
• Find for bugs running the app on iOS

1. Write JavaDoc comments for all class variables and functions.
2. In many cases, there are multi-line JavaDoc comments where the opening /** and closing */ could be moved onto the same line as the comment itself.
3. There are functions with unfinished JavaDoc comments.
4. JavaDoc comments on classes can be removed if they have no useful information (Ex. A one-line description of t