Automatic Speech Recognition
NeMo
PyTorch
Bambara
speech
audio
CTC
QuartzNet
legacy-model
deprecated
Bambara
NeMo
Eval Results (legacy)
Instructions to use RobotsMali/anbekalanNet with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- NeMo
How to use RobotsMali/anbekalanNet with NeMo:
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.ASRModel.from_pretrained("RobotsMali/anbekalanNet") transcriptions = asr_model.transcribe(["file.wav"]) - Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -130,50 +130,10 @@ The performance metrics below illustrate how expanding data volume rescued the Q
|
|
| 130 |
|
| 131 |
| Experimental Pass | Dataset Baseline Configuration | SpecAugment | Training Volume | Test WER (%) ↓ | Test CER (%) ↓ |
|
| 132 |
| --- | --- | --- | --- | --- | --- |
|
| 133 |
-
| **anbekalanNet-exp3 (this release)** | <br>**Main + Duplicate**
|
| 134 |
-
|
| 135 |
-
| <br>**
|
| 136 |
-
|
| 137 |
-
| <br>**45.6 Hours**
|
| 138 |
-
|
| 139 |
-
| <br>**40.0%**
|
| 140 |
-
|
| 141 |
-
| <br>**15.0%**
|
| 142 |
-
|
| 143 |
-
|
|
| 144 |
-
| *anbekalanNet-exp1* | <br>*Main Only*
|
| 145 |
-
|
| 146 |
-
| <br>*None*
|
| 147 |
-
|
| 148 |
-
| <br>*1.6 Hours*
|
| 149 |
-
|
| 150 |
-
| <br>*93.0%*
|
| 151 |
-
|
| 152 |
-
| <br>*80.0%*
|
| 153 |
-
|
| 154 |
-
|
|
| 155 |
-
| *anbekalanNet-exp2* | <br>*Main Only*
|
| 156 |
-
|
| 157 |
-
| <br>*Active*
|
| 158 |
-
|
| 159 |
-
| <br>*1.6 Hours*
|
| 160 |
-
|
| 161 |
-
| <br>*64.0%*
|
| 162 |
-
|
| 163 |
-
| <br>*23.0%*
|
| 164 |
-
|
| 165 |
-
|
|
| 166 |
-
| *anbekalanNet-exp4* | <br>*Main + Duplicate*
|
| 167 |
-
|
| 168 |
-
| <br>*Active*
|
| 169 |
-
|
| 170 |
-
| <br>*45.6 Hours*
|
| 171 |
-
|
| 172 |
-
| <br>*42.0%*
|
| 173 |
-
|
| 174 |
-
| <br>*16.0%*
|
| 175 |
-
|
| 176 |
-
|
|
| 177 |
|
| 178 |
*All results indicate greedy decoding performance without external Language Models (LMs).*
|
| 179 |
|
|
|
|
| 130 |
|
| 131 |
| Experimental Pass | Dataset Baseline Configuration | SpecAugment | Training Volume | Test WER (%) ↓ | Test CER (%) ↓ |
|
| 132 |
| --- | --- | --- | --- | --- | --- |
|
| 133 |
+
| **anbekalanNet-exp3 (this release)** | <br>**Main + Duplicate** | <br>**None** | <br>**45.6 Hours** | <br>**40.0%** | <br>**15.0%** |
|
| 134 |
+
| *anbekalanNet-exp1* | <br>*Main Only* | <br>*None* | <br>*1.6 Hours* | <br>*93.0%* | <br>*80.0%* |
|
| 135 |
+
| *anbekalanNet-exp2* | <br>*Main Only* | <br>*Active* | <br>*1.6 Hours* | <br>*64.0%* | <br>*23.0%* |
|
| 136 |
+
| *anbekalanNet-exp4* | <br>*Main + Duplicate* | <br>*Active* | <br>*45.6 Hours* | <br>*42.0%* | <br>*16.0%* |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
*All results indicate greedy decoding performance without external Language Models (LMs).*
|
| 139 |
|