Card: exact size-class wording
Browse files
README.md
CHANGED
|
@@ -36,8 +36,8 @@ Devanagari and English words stay in English, the way people actually type: *उ
|
|
| 36 |
It runs on a phone, a laptop or inside a web page, with no server and no internet connection. On a 2021 Samsung Galaxy
|
| 37 |
F62 it transcribes 12x faster than real time using 64 MB of memory.
|
| 38 |
|
| 39 |
-
Of the open speech-to-text models under 100 MB that we measured, it is the most accurate on Marathi
|
| 40 |
-
|
| 41 |
[Shruti Hindi](https://huggingface.co/udgamlabs/Shruti), and is built on [Cactus Whistle](https://huggingface.co/Cactus-Compute/whistle)
|
| 42 |
(Apache-2.0) via Shruti Hindi: we retrained all its weights on openly licensed Marathi and English speech and added 400 Marathi
|
| 43 |
vocabulary pieces.
|
|
|
|
| 36 |
It runs on a phone, a laptop or inside a web page, with no server and no internet connection. On a 2021 Samsung Galaxy
|
| 37 |
F62 it transcribes 12x faster than real time using 64 MB of memory.
|
| 38 |
|
| 39 |
+
Of the open speech-to-text models under 100 MB that we measured, it is the most accurate on Marathi. IndicConformer
|
| 40 |
+
Marathi, which is more accurate on read Marathi, is about 5x larger. Shruti Marathi is the Marathi member of the Shruti family, next to
|
| 41 |
[Shruti Hindi](https://huggingface.co/udgamlabs/Shruti), and is built on [Cactus Whistle](https://huggingface.co/Cactus-Compute/whistle)
|
| 42 |
(Apache-2.0) via Shruti Hindi: we retrained all its weights on openly licensed Marathi and English speech and added 400 Marathi
|
| 43 |
vocabulary pieces.
|