19 Commits

Author	SHA1	Message	Date
marcoyang1998	81af525de4	update the biasing lists	2023-09-08 10:15:21 +08:00
marcoyang1998	bbf1577818	add long audio transcription scripts	2023-09-08 10:02:41 +08:00
marcoyang1998	07e27348dd	more updates	2023-09-08 10:01:48 +08:00
marcoyang1998	013cafdd6d	updates	2023-09-08 10:00:00 +08:00
marcoyang1998	522273f97e	change the text normalization for upper_case_no_punc	2023-09-08 09:57:24 +08:00
marcoyang1998	77890a6115	add context biasing at different levels	2023-09-08 09:56:45 +08:00
marcoyang1998	d4c5a1c157	updates	2023-09-08 09:55:41 +08:00
marcoyang1998	cad01bfcb6	add subformer model with style embeddings	2023-08-29 16:04:51 +08:00
marcoyang1998	16e8907805	update text normalization for librispeech test sets	2023-08-29 16:03:56 +08:00
marcoyang1998	80c54c05e2	support showing WERs of different books	2023-08-17 23:59:37 +08:00
marcoyang1998	f23882b9f6	also sample from distractors when using separate words in the ref text; increase the max length of substring	2023-08-17 12:11:33 +08:00
marcoyang1998	8a238317a4	support using subformer as text encoder and train with style	2023-08-16 19:08:36 +08:00
marcoyang1998	73fa1651f0	minor updates to utils.py	2023-08-16 16:47:23 +08:00
marcoyang1998	2091bb5f25	add two pass decoding	2023-08-16 16:46:50 +08:00
marcoyang1998	0982db9cde	add a few args to support context list and rare words	2023-08-16 16:44:58 +08:00
marcoyang1998	4420788f66	support using context list and random substring as pre text	2023-08-16 16:44:29 +08:00
marcoyang1998	17d0918969	fix the post normalization bug, avoid multiple words	2023-08-16 09:39:42 +08:00
marcoyang1998	fdc4fcabb9	use a more aggresive sampling_weight	2023-08-16 09:38:40 +08:00
marcoyang1998	ae4d2fbfcc	initial commit	2023-08-14 09:51:20 +08:00