Update README.md
Browse files
README.md
CHANGED
@@ -16,6 +16,7 @@ pipeline_tag: automatic-speech-recognition
|
|
16 |
|
17 |
**This is a CTC-based Automatic Speech Recognition system for French.**
|
18 |
This model is part of the SLU demo available here: [LINK TO THE DEMO GOES HERE]
|
|
|
19 |
It is based on the [mHuBERT-147](https://huggingface.co/utter-project/mHuBERT-147) speech foundation model.
|
20 |
|
21 |
* Training data: XX hours
|
@@ -24,20 +25,21 @@ It is based on the [mHuBERT-147](https://huggingface.co/utter-project/mHuBERT-14
|
|
24 |
|
25 |
|
26 |
# Table of Contents:
|
27 |
-
1. Training Parameters
|
28 |
-
2. [ASR Model class](https://huggingface.co/naver/mHuBERT-147-ASR-fr#ASR-Model-
|
29 |
-
3. Running inference
|
30 |
|
31 |
## Training Parameters
|
32 |
-
The training parameters are available in config.yaml.
|
|
|
33 |
|
34 |
-
## ASR Model
|
35 |
|
36 |
We use the mHubertForCTC class for our model, which is nearly identical to the existing HubertForCTC class.
|
37 |
The key difference is that we've added a few additional hidden layers at the end of the Transformer stack, just before the lm_head.
|
38 |
The code is available in [CTC_model.py](https://huggingface.co/naver/mHuBERT-147-ASR-fr/blob/main/CTC_model.py).
|
39 |
|
40 |
-
## Running
|
41 |
|
42 |
The run_asr.py file illustrates how to load the model for inference (**load_asr_model**), and how to produce transcription for a file (**run_asr_inference**).
|
43 |
Please follow the [requirements file](https://huggingface.co/naver/mHuBERT-147-ASR-fr/blob/main/requirements.txt) to avoid incorrect model loading.
|
|
|
16 |
|
17 |
**This is a CTC-based Automatic Speech Recognition system for French.**
|
18 |
This model is part of the SLU demo available here: [LINK TO THE DEMO GOES HERE]
|
19 |
+
|
20 |
It is based on the [mHuBERT-147](https://huggingface.co/utter-project/mHuBERT-147) speech foundation model.
|
21 |
|
22 |
* Training data: XX hours
|
|
|
25 |
|
26 |
|
27 |
# Table of Contents:
|
28 |
+
1. [Training Parameters](https://huggingface.co/naver/mHuBERT-147-ASR-fr#Training-Parameters)
|
29 |
+
2. [ASR Model class](https://huggingface.co/naver/mHuBERT-147-ASR-fr#ASR-Model-Class)
|
30 |
+
3. [Running inference](https://huggingface.co/naver/mHuBERT-147-ASR-fr#Running-Inference)
|
31 |
|
32 |
## Training Parameters
|
33 |
+
The training parameters are available in [config.yaml](https://huggingface.co/naver/mHuBERT-147-ASR-fr/blob/main/config.yaml).
|
34 |
+
We highlight the use of 0.3 for hubert.final_dropout, which we found to be very helpful in convergence. We also use fp32 training, as we found fp16 training to be unstable.
|
35 |
|
36 |
+
## ASR Model Class
|
37 |
|
38 |
We use the mHubertForCTC class for our model, which is nearly identical to the existing HubertForCTC class.
|
39 |
The key difference is that we've added a few additional hidden layers at the end of the Transformer stack, just before the lm_head.
|
40 |
The code is available in [CTC_model.py](https://huggingface.co/naver/mHuBERT-147-ASR-fr/blob/main/CTC_model.py).
|
41 |
|
42 |
+
## Running Inference
|
43 |
|
44 |
The run_asr.py file illustrates how to load the model for inference (**load_asr_model**), and how to produce transcription for a file (**run_asr_inference**).
|
45 |
Please follow the [requirements file](https://huggingface.co/naver/mHuBERT-147-ASR-fr/blob/main/requirements.txt) to avoid incorrect model loading.
|