🚀 update training results for contextnet

TensorSpeech · Jun 10, 2021 · 130124c · 130124c
1 parent c6eeb03
commit 130124c
Show file tree

Hide file tree

Showing 3 changed files with 30 additions and 0 deletions.
diff --git a/examples/contextnet/README.md b/examples/contextnet/README.md
@@ -17,3 +17,31 @@ Training, see `python examples/contextnet/train_*.py --help`
 Testing, see `python examples/contextnet/test_*.py --help`
 
 TFLite Conversion, see `python examples/contextnet/tflite_*.py --help`
+
+## RNN Transducer Subwords - Results on LibriSpeech
+
+**Summary**
+
+- Number of subwords: 1008
+- Maximum length of a subword: 10
+- Subwords corpus: all training sets
+- Number of parameters: 12,075,320
+- Number of epochs: 86
+- Train on: 8 Google Colab TPUs
+- Train hours: 8.375 days uncontinuous (each day I trained 12 epoch because colab only allows 12 hours/day and 1 epoch required 1 hour) => 86 hours continuous (3.58333333 days)
+
+**Pretrained and Config**, go to [drive](https://drive.google.com/drive/folders/1fzOkwKaOcMUMD9BAjcLLmSG2Tfpeabbq?usp=sharing)
+
+**Epoch Transducer Loss**
+
+<img src="./figs/1008_subword_contextnet_loss.svg" alt="subword_contextnet_loss" width="300px" />
+
+**Epoch Learning Rate**
+
+<img src="./figs/1008_epoch_learning_rate.svg" alt="epoch_learning_rate" width="300px" />
+
+**Error Rates**
+
+| **Test-clean** | Test batch size | Epoch |      WER (%)       |      CER (%)       |
+| :------------: | :-------------: | :---: | :----------------: | :----------------: |
+|    _Greedy_    |        1        |  86   | 10.356436669826508 | 5.8370333164930344 |
diff --git a/examples/contextnet/figs/1008_epoch_learning_rate.svg b/examples/contextnet/figs/1008_epoch_learning_rate.svg