Commit
•
baba7fe
1
Parent(s):
26c5685
Update README.md
Browse files
README.md
CHANGED
|
@@ -5,7 +5,6 @@ language:
|
|
| 5 |
pipeline_tag: automatic-speech-recognition
|
| 6 |
library_name: nemo
|
| 7 |
---
|
| 8 |
-
|
| 9 |
## IndicConformer
|
| 10 |
|
| 11 |
IndicConformer is a Hybrid RNNT conformer model built for Hindi.
|
|
@@ -22,7 +21,6 @@ library_name: nemo
|
|
| 22 |
```bash
|
| 23 |
$ python inference.py --help
|
| 24 |
usage: inference.py [-h] -c CHECKPOINT -f AUDIO_FILEPATH -d (cpu,cuda) -l LANGUAGE_CODE
|
| 25 |
-
|
| 26 |
options:
|
| 27 |
-h, --help show this help message and exit
|
| 28 |
-c CHECKPOINT, --checkpoint CHECKPOINT
|
|
@@ -37,7 +35,7 @@ library_name: nemo
|
|
| 37 |
|
| 38 |
## Example command
|
| 39 |
```
|
| 40 |
-
python inference.py -c
|
| 41 |
```
|
| 42 |
Expected output -
|
| 43 |
|
|
@@ -61,33 +59,5 @@ library_name: nemo
|
|
| 61 |
|
| 62 |
## Model Architecture
|
| 63 |
|
| 64 |
-
This model is a conformer-Large model, consisting of 120M parameters, as the encoder, with a hybrid CTC-RNNT
|
| 65 |
512 as the model dimension.
|
| 66 |
-
|
| 67 |
-
## Training
|
| 68 |
-
|
| 69 |
-
<ADD INFORMATION ABOUT HOW THE MODEL WAS TRAINED - HOW MANY EPOCHS, AMOUNT OF COMPUTE ETC>
|
| 70 |
-
|
| 71 |
-
### Datasets
|
| 72 |
-
|
| 73 |
-
<LIST THE NAME AND SPLITS OF DATASETS USED TO TRAIN THIS MODEL (ALONG WITH LANGUAGE AND ANY ADDITIONAL INFORMATION)>
|
| 74 |
-
|
| 75 |
-
## Performance
|
| 76 |
-
|
| 77 |
-
<LIST THE SCORES OF THE MODEL -
|
| 78 |
-
OR
|
| 79 |
-
USE THE Hugging Face Evaluate LiBRARY TO UPLOAD METRICS>
|
| 80 |
-
|
| 81 |
-
## Limitations
|
| 82 |
-
|
| 83 |
-
<DECLARE ANY POTENTIAL LIMITATIONS OF THE MODEL>
|
| 84 |
-
|
| 85 |
-
Eg:
|
| 86 |
-
Since this model was trained on publicly available speech datasets, the performance of this model might degrade for speech which includes technical terms, or vernacular that the model has not been trained on. The model might also perform worse for accented speech.
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
## References
|
| 90 |
-
|
| 91 |
-
<ADD ANY REFERENCES HERE AS NEEDED>
|
| 92 |
-
|
| 93 |
-
[1] [AI4Bharat NeMo Toolkit](https://github.com/AI4Bharat/NeMo)
|
|
|
|
| 5 |
pipeline_tag: automatic-speech-recognition
|
| 6 |
library_name: nemo
|
| 7 |
---
|
|
|
|
| 8 |
## IndicConformer
|
| 9 |
|
| 10 |
IndicConformer is a Hybrid RNNT conformer model built for Hindi.
|
|
|
|
| 21 |
```bash
|
| 22 |
$ python inference.py --help
|
| 23 |
usage: inference.py [-h] -c CHECKPOINT -f AUDIO_FILEPATH -d (cpu,cuda) -l LANGUAGE_CODE
|
|
|
|
| 24 |
options:
|
| 25 |
-h, --help show this help message and exit
|
| 26 |
-c CHECKPOINT, --checkpoint CHECKPOINT
|
|
|
|
| 35 |
|
| 36 |
## Example command
|
| 37 |
```
|
| 38 |
+
python inference.py -c indicconformer_stt_hi_hybrid_rnnt_large.nemo -f hindi-16khz.wav -d cuda -l hi
|
| 39 |
```
|
| 40 |
Expected output -
|
| 41 |
|
|
|
|
| 59 |
|
| 60 |
## Model Architecture
|
| 61 |
|
| 62 |
+
This model is a conformer-Large model, consisting of 120M parameters, as the encoder, with a hybrid CTC-RNNT decoder. The model has 17 conformer blocks with
|
| 63 |
512 as the model dimension.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|