[hts-users:04465] Re: get bad voice by run HTS-2.3.1

[Niya N <niya121212@xxxxxxxxx>]

how to generate bap files . Please help

On Mon, Feb 6, 2017 at 2:34 PM, IMUCS_liurui <liurui_imu@xxxxxxx> wrote:

OK! I try to get more data to test.

thank you very much!

在 2017-02-06 17:02:58，"Bùi Tấn Quang" <quangxengvn@xxxxxxxxx> 写道：

... it depends on your data !

in your current state, it's true

2017-02-06 15:50 GMT+07:00 IMUCS_liurui <liurui_imu@xxxxxxx>:

Thank you!

Will increasing the training data can increase the naturalness of the results?

在 2017-02-06 16:48:39，"Bùi Tấn Quang" <quangxengvn@xxxxxxxxx> 写道：

it is too little

2017-02-06 15:43 GMT+07:00 IMUCS_liurui <liurui_imu@xxxxxxx>:

298utt.

在 2017-02-06 16:42:36，"Bùi Tấn Quang" <quangxengvn@xxxxxxxxx> 写道：

298 utt? 

298 MB?

298 hours?

2017-02-06 15:41 GMT+07:00 IMUCS_liurui <liurui_imu@xxxxxxx>:

my train data size is 298
，too little training data?

在 2017-02-06 16:36:16，"Bùi Tấn Quang" <quangxengvn@xxxxxxxxx> 写道：

i think it only show you the way to build voice with DNN 

2017-02-05 14:35 GMT+07:00 IMUCS_liurui <liurui_imu@xxxxxxx>:

I used the same data to run 2.3 and 2.3.1, but found the results of 2.3.1（dnn）is no better than 2.3

excuse me, can you tell me the reason?

Thank everyone.:)