Heejo You comments

Results 8 comments of


                                            Heejo You

confusion with speaker encoder and loss func

@arunasank As I understand L_recon is `after postnet mel` and L_recon0 is `pre postnet mel`. Tacotron 2 used the similar postnet structure and loss.

confusion with speaker encoder and loss func

@Trebolium The Tacotron2 paper does not specifically mention the purpose of Postnet. My personal guess is that it increases the detail of the mel. Postnet is a residual structure(_postnet =...

confusion with speaker encoder and loss func

@ruclion Interesting. Now I don't know the clear reason. I think it would be better to ask the author of the paper (the owner of this repository).... :) And, please...

pre trained model

Hi, the shape difference of inference mels are because of the dropout in prenet. I think you try to generate the mels one more time, the shape may be different...

The mismatch of sampling rate of Glow-TTS acoustic model and the vocoder

Dear @actuy, Hello. I apologize for the delayed response. After the issue was raised, I searched the correct vocoder and checkpoint. Although I couldn't find a version updated to 24K...

One-hot code speaker embedding and GE2E speaker embedding.

Hi, rishabhjain16, Thank you for contacting me. Unfortunately, this project has been stopped due to issues with the conversion quality. Currently implemented models do not use embedding. Therefore, a vector...

[Performance] The implementation of make_block_ptr matmul appears to have worse performance than the official tutorial implementation.

Dear @tlogn , Hello, I checked your code and found an issue in one part. ``` a_tile = tl.make_block_ptr(a_ptr, shape=(M, K), strides=(K, 1), offsets=(pid_m*BLOCK_M, 0), block_shape=(BLOCK_M, BLOCK_K), order=(1, 0)) b_tile...

[Performance] The implementation of make_block_ptr matmul appears to have worse performance than the official tutorial implementation.

Hi, here is my code: ``` pid = tl.program_id(axis=0) num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) num_pid_in_group = GROUP_SIZE_M * num_pid_n group_id = pid // num_pid_in_group first_pid_m = group_id...