eval/validation fix for SpeechX tasks
This commit is contained in:
parent
28a674e0f1
commit
692d09f9c1
|
@ -99,6 +99,10 @@ def run_eval(engines, eval_name, dl):
|
||||||
if task != "tts":
|
if task != "tts":
|
||||||
filename = f"{filename}_{task}"
|
filename = f"{filename}_{task}"
|
||||||
|
|
||||||
|
# flatten prom
|
||||||
|
if not isinstance(prom, torch.Tensor):
|
||||||
|
prom = torch.concat([ p for p in prom if isinstance(p, torch.Tensor) ])
|
||||||
|
|
||||||
# to-do, refine the output dir to be sane-er
|
# to-do, refine the output dir to be sane-er
|
||||||
ref_path = (cfg.log_dir / str(engines.global_step) / "ref" / filename).with_suffix(".wav")
|
ref_path = (cfg.log_dir / str(engines.global_step) / "ref" / filename).with_suffix(".wav")
|
||||||
hyp_path = (cfg.log_dir / str(engines.global_step) / name / eval_name / filename).with_suffix(".wav")
|
hyp_path = (cfg.log_dir / str(engines.global_step) / name / eval_name / filename).with_suffix(".wav")
|
||||||
|
|
Loading…
Reference in New Issue
Block a user