clean commit with EMTTS text embedding disabling and pronunciation co…#15717
clean commit with EMTTS text embedding disabling and pronunciation co…#15717paarthneekhara wants to merge 7 commits into
Conversation
blisc
left a comment
blisc
left a comment
There was a problem hiding this comment.
Let's add more documentation to what these changes are.
| from nemo.utils import logging | ||
|
|
||
|
|
||
| def setup_pronunciation_control_g2p(pronunciation_control_g2p_config): |
There was a problem hiding this comment.
Can you add docstrings to your new functions?
| return aggregated_tokenizer | ||
|
|
||
|
|
||
| def setup_pronunciation_control_g2p(pronunciation_control_g2p_config): |
There was a problem hiding this comment.
If the new functions are the same between the two datasets, can we merge the two functions and import rather than defining it twice?
There was a problem hiding this comment.
Moved them to tts_dataset_utils.py and added docstrings.
| self.phoneme_loss_weight = cfg.get('phoneme_loss_weight', 1.0) | ||
| self.parallel_codebook_loss_scale = cfg.get('parallel_codebook_loss_scale', 1.0) | ||
| self.local_transformer_loss_scale = cfg.get('local_transformer_loss_scale', 1.0) | ||
| self.phoneme_as_text_prob = cfg.get('phoneme_as_text_prob', 0.0) |
There was a problem hiding this comment.
Where is this new parameter being defined? Or is it always going to be 0?
There was a problem hiding this comment.
Added this in the yamls. In the candidate models it is 0. It is non zero for pronunciation control experiments.
| self.disable_subword_embedding = cfg.get('disable_subword_embedding', False) | ||
| self.disable_lm_text_head = cfg.get('disable_lm_text_head', False) | ||
| if self.disable_subword_embedding and not self.use_bpe_char_tokenizer: | ||
| logging.warning( | ||
| "`disable_subword_embedding=True` requires `use_bpe_char_tokenizer=True`; overriding automatically." | ||
| ) | ||
| self.use_bpe_char_tokenizer = True |
There was a problem hiding this comment.
Can you document what your new arguments mean?
There was a problem hiding this comment.
Added comments about these arguments.
…ntrol changes Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
paarthneekhara
force-pushed
the
emtts_disabletextembedding
branch
from
ac983a9 to
c9a7c40
Compare
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
|
/ok to test ff77b14 |
Signed-off-by: Paarth Neekhara <paarth.n@gmail.com>
|
/ok to test a6f7bd2 |
2 participants
No description provided.