method

[yunzqq]

it is operate on TF domain right?

[JasonSWFu]

Yes

[yunzqq]

Yes

Thks

[yunzqq]

For doing comparison in the voice bank + DEMAND dataset, i want to know the parameter size of Metric GAN (G or D each rescpectively) thks very much!

[yunzqq]

Dear Szu-Wei Fu

I am a student from harbin institute of technology. I am doing research on speech enhancement.

Recently, i am reading your paper Metric GAN speech enhancement.

I want to consult some issues (1) how many parameters for G and D networks, respectively? (2) G network use BLSTM netowork, so it is a non-causal system, right?

Thank you very much for your help! I am looking forward to your reply.

Best, Qiquan Zhang, HIT

[email protected]

From: Szu-Wei Fu Date: 2019-11-07 15:17 To: JasonSWFu/MetricGAN CC: George_Zhang; Author Subject: Re: [JasonSWFu/MetricGAN] method (#5) Yes — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[JasonSWFu]

Hi Qiquan,

1. After checking model.summary(), the number of parameters for G and D are 1,892,057 and 690,855, respectively.
2. Yes, it is non-causal.

Sincerely, Szu-Wei

[yunzqq]

Hi Qiquan,

1. After checking model.summary(), the number of parameters for G and D are 1,892,057 and 690,855, respectively.
2. Yes, it is non-causal.

Sincerely, Szu-Wei

Hi, Szu-Wei Thanks, it is same as i check, thank you !

Sincerely, Qiquan

[yunzqq]

and part of parameters in D network are not learnable.

[yunzqq]

Dear Szu-Wei Fu,

We are making comparison with MetricGAN on the DEMAND database. For results of metrics, we use the value reported in the paper. But, we want to do a comparison on spectrograms of speech. So, could you supply the enhanced speech or pre-trained model for us?

Thank you very much for your help！

Best, Qiquan Zhang Harbin Institute of Technology

[email protected]

From: Szu-Wei Fu Date: 2019-12-02 14:14 To: JasonSWFu/MetricGAN CC: George_Zhang; Author Subject: Re: [JasonSWFu/MetricGAN] method (#5) Hi Qiquan, After checking model.summary(), the number of parameters for G and D are 1,892,057 and 690,855, respectively. Yes, it is non-causal. Sincerely, Szu-Wei — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.