-
Conceptually replicating the analysis of our Interspeech paper is a useful goal to guide scikit-talk development. To that end I'm going to create a proof of concept of our code for identifying continu…
-
## 🐛 Bug
Regard using the DNS-Challenge baseline, there are issues when using the newest version of the DNS Challenge repo, where the main problems occur when evaluating models, as the testdata has b…
-
* [IEEE ICASSP](https://2021.ieeeicassp.org/): International Conference on Acoustics, Speech and Signal Processing
* Deadline already passed in October
* [ACM Multimedia](https://2021.acmmm.org/…
hofbi updated
2 years ago
-
i install ed
![image](https://user-images.githubusercontent.com/11546854/44587514-c8480780-a7d0-11e8-9416-8d1387c915a8.png)
but when i tryied to run
![image](https://user-images.githubuserconten…
-
下载了您的代码,只修改了数据,数据用的是interspeech PLC竞赛的数据,训了13个epoch,丢帧生成的数据几乎为0.
-
Thank for your sharing!
Could you tell me the name of article that you refer to in this code?
Looking forward to your reply/
-
-
I am trying to continue training the pre-trained FullSubNet model provided by this repo:
[fullsubnet_best_model_58epochs.tar](https://github.com/haoxiangsnr/FullSubNet/releases/download/v0.2/fullsu…
-
Hello!
I'm recently doing some experiments on NMSQA, and the code for DUAL provided here are really helpful! While I encountered some difficulty building the units using scripts provided to reprod…
-
Hi. I ran this project for korean speech recognition.
But loss is not decreasing and i don't get good predictions.
I've already used preprocessing method that works well on DeepSpeech and LAS.
It…