Jong Wook Kim
6dea21fd7f
Release 20230314
2023-03-15 00:39:19 -07:00
Jong Wook Kim
ad3250a846
Release 20230308
2023-03-08 15:48:57 -08:00
Jong Wook Kim
aac47c9834
fix typo
2023-03-07 20:43:49 -08:00
Jong Wook Kim
26807ec6d3
Release 20230307
2023-03-07 20:36:29 -08:00
Jong Wook Kim
8180fde939
Release 20230306
2023-03-06 18:53:04 -08:00
Jong Wook Kim
55f690af79
Release 20230124
2023-01-24 11:11:08 -08:00
Jong Wook Kim
8135a7c31c
verbose outputs from pytest
2023-01-18 10:30:18 -08:00
Jong Wook Kim and GitHub
9d646db9d8
print '?' if a letter can't be encoded using the system default encoding ( #859 )
2023-01-17 23:28:36 -08:00
Jong Wook Kim
37a4f1be6d
Release 20230117
2023-01-17 16:08:28 -08:00
Jong Wook Kim
a84191faae
rename GitHub workflow
2023-01-17 13:54:40 -08:00
Jong Wook Kim
b1d213c0c7
allow test_transcribe to run on CPU when CUDA is not available
2023-01-17 13:43:36 -08:00
Jong Wook Kim
493dfffa37
add github action to run pytest
2023-01-17 13:38:33 -08:00
Jong Wook Kim
f82bc59f5e
torch.concatenate -> torch.cat for compatibility
2023-01-10 10:53:18 -08:00
Jong Wook Kim
28769fcfe5
word-level timestamps in Multilingual_ASR notebook
2022-12-31 10:03:42 -07:00
Jong Wook Kim
53807677fe
MultiHeadAttention to return qk as well
2022-12-30 01:53:57 -07:00
Jong Wook Kim
9323b2526c
Revert "saving the qk matrix in the attention module for convenience"
...
This reverts commit 68e44bd83c
2022-12-29 23:53:31 -07:00
Jong Wook Kim
68e44bd83c
saving the qk matrix in the attention module for convenience
2022-12-29 23:02:52 -07:00
Jong Wook Kim
0b5dcfdef7
large-v2 figure and arxiv url update
2022-12-09 00:12:39 -05:00
Jong Wook Kim
4179ed2475
add large-v2 model
...
- The "large-v2" model is trained for more epochs with regularization and shows improved performance compared to the previous large.
- It has the same architecture as the original large model.
- When `load_model("large")` is called, the "large-v2" model will be loaded.
- We will soon update the paper regarding this new model.
2022-12-05 11:07:14 -05:00
Jong Wook Kim
eff383b27b
invoking __call__ instead of forward()
2022-11-16 04:18:50 -08:00
Jong Wook Kim
02aa851a49
fix to return only the text token ids
2022-11-15 16:25:11 -08:00
Jong Wook Kim
d18e9ea5dd
transcribe() on English-only model won't complain when language="en" is not given
2022-10-09 02:40:12 -07:00
Jong Wook Kim
7cb4cc21bf
allowing nonzero initial temperature
2022-09-29 18:05:12 -07:00
Jong Wook Kim
30dc5c581b
pointer to the show and tell section
2022-09-29 14:57:49 -07:00
Jong Wook Kim
62fe7f1009
patience definition to match the paper
2022-09-27 19:00:41 -07:00
Jong Wook Kim
5d8d3e75a4
add --condition_on_previous_text
2022-09-25 05:16:08 -07:00
Jong Wook Kim
2d3032de01
improved warning message for English-only models
2022-09-25 02:10:36 -07:00
Jong Wook Kim
8cf36f3508
allow hyphens and single quotes between words
2022-09-23 20:11:27 +09:00
Jong Wook Kim
15ab548263
nocaptions -> nospeech to match the paper figure
2022-09-23 15:45:32 +09:00
Jong Wook Kim and GitHub
e90b8fa7e8
Merge pull request #14 from bquast/patch-1
...
make LICENSE a link instead of code-formatted text
2022-09-22 11:51:05 +09:00
Jong Wook Kim and GitHub
f83cb83a42
Merge pull request #24 from ldanilov/patch-1
...
fixes the link to the model paper
2022-09-22 11:48:57 +09:00
Jong Wook Kim
49a3ffc997
add section Available models and languages
2022-09-22 05:36:25 +09:00
Jong Wook Kim
cfd6bdda21
a note on speed-accuracy tradeoffs
2022-09-22 02:58:56 +09:00
Jong Wook Kim
834f00a0ea
making small model the default
2022-09-22 02:45:12 +09:00
Jong Wook Kim
6e3be77e1a
initial commit
2022-09-22 01:09:43 +09:00
Jong Wook Kim
