Added training instructions to README.md

MiscellaneousStuff · May 5, 2022 · 93b3f84 · 93b3f84
1 parent e56304c
commit 93b3f84
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 6 deletions.
diff --git a/README.md b/README.md
@@ -53,6 +53,16 @@ git submodule init
 git submodule update
 ```
 
+## Train
+
+To train an sEMG Silent Speech speech recognition model, use
+
+```bash
+python3 train.py \
+ --dataset_path "path_to_dataset.csv" \
+ --semg_train
+```
+
 ## Evaluate
 
 To evaluate the best trained model released with the report, run the
@@ -62,6 +72,7 @@ following code:
 python3 evaluate.py \
  --checkpoint_path "path_to_pretrained_model/ds2_DATASET_SILENT_SPEECH_EPOCHS_10_TEST_LOSS_1.8498832106590273_WER_0.6825681123095443" \
  --dataset_path "path_to_dataset.csv" \
+ --print_top 10 \
  --semg_eval
 ```
 

diff --git a/evaluate.py b/evaluate.py
@@ -44,10 +44,10 @@
 FLAGS = flags.FLAGS
 flags.DEFINE_string("checkpoint_path", None, "Path to the pre-trained DeepSpeech2 model")
 flags.DEFINE_boolean("semg_eval", False, \
- "(Optional) Evaluate an ASR model on predicted mel spectrograms from the transducer."
+ "(Optional) Evaluate an ASR model on predicted mel spectrograms from the transducer.\n"
  "Otherwise evaluate the ground truth audio files.")
 flags.DEFINE_integer("random_seed", 7, \
- "(Optional) Set a different random seed if you train a different model."
+ "(Optional) Set a different random seed if you train a different model.\n"
  "The models trained along with this release used a random seed of 7 by default.")
 flags.DEFINE_string("dataset_path", None, \
  "Path to *.csv file which defines the dataset to evaluate")

diff --git a/train.py b/train.py
@@ -50,16 +50,16 @@
 flags.DEFINE_integer("n_epochs", 100,\
  "Recommended epochs is 200 for closed vocab and 100 for the others")
 flags.DEFINE_integer("random_seed", 7, \
- "(Optional) Set a different random seed if you train a different model."
+ "(Optional) Set a different random seed if you train a different model.\n"
  "The models trained along with this release used a random seed of 7 by default.")
 flags.DEFINE_boolean("semg_train", False, \
- "(Optional) Train an ASR model on predicted mel spectrograms from the transducer."
+ "(Optional) Train an ASR model on predicted mel spectrograms from the transducer.\n"
  "Otherwise train on the ground truth audio files.")
 flags.DEFINE_boolean("silent_only", False, \
- "(Optional) Transduction dataset only."
+ "(Optional) Transduction dataset only.\n"
  "Train only on the mel spectrograms predicted from EMG signals during silent speech")
 flags.DEFINE_boolean("voiced_only", False, \
- "(Optional) Transduction dataset only."
+ "(Optional) Transduction dataset only.\n"
  "Train only on the mel spectrograms predicted from EMG signals during vocalised speech")
 flags.DEFINE_boolean("amp", False, \
  "(Optional) Train using Automatic Mixed Precision (AMP)")