mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

tweaked tts-continuous task

Browse Source
This commit is contained in:
mrq 2023-09-02 13:39:17 -05:00
parent 21e5d250cc
commit 71e68a8528
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vall_e/data.py
View File

@ -314,12 +314,14 @@ class Dataset(_Dataset):
# text-to-speech # text-to-speech
if task == "tts" or task == "tts-c": if task == "tts" or task == "tts-c":
trim_length = int(cfg.dataset.prompt_duration * 75) trim_length = int(cfg.dataset.prompt_duration * 75)
continuous = task == "tts-c" and trim_length * 2 < resps.shape[0] # demote if the target is too short
if task == "tts-c" and trim_length * 2 >= resps.shape[0]:
task = "tts"
# VALL-E continuous # VALL-E continuous
# ignore if target utterance is shorter than prompt duration # ignore if target utterance is shorter than prompt duration
# to-do: actually do this for the AR only as I don't think the paper trained the NAR for this # to-do: actually do this for the AR only as I don't think the paper trained the NAR for this
if continuous: if task == "tts-c":
proms = resps[:trim_length, :] proms = resps[:trim_length, :]
resps = resps[trim_length:, :] resps = resps[trim_length:, :]
else: else: