[update] process original ds

train2.py

@@ -1,56 +1,41 @@
-from datasets import load_dataset, DatasetDict
-from datasets import Audio
+import librosa
+import numpy as np
+from datasets import Audio, DatasetDict, load_dataset
 
+from training import utils
 from training.train import Trainer
 
-common_voice = DatasetDict()
-
-common_voice["train"] = load_dataset(
- "mozilla-foundation/common_voice_11_0",
- "hi",
- split="train+validation",
- use_auth_token=True,
-)
-common_voice["test"] = load_dataset(
- "mozilla-foundation/common_voice_11_0", "hi", split="test", use_auth_token=True
-)
-
-
-common_voice = common_voice.remove_columns(
- [
- "accent",
- "age",
- "client_id",
- "down_votes",
- "gender",
- "locale",
- "path",
- "segment",
- "up_votes",
- ]
-)
-
-common_voice = common_voice.cast_column("audio", Audio(sampling_rate=16000))
+DS_PATH = "dataset/"
 
+dataset = utils.gather_dataset(DS_PATH)
 trainer = Trainer()
 
+is_prepared = False
 
-def prepare_dataset(batch):
- # load and resample audio data from 48 to 16kHz
- audio = batch["audio"]
+if not is_prepared:
+ target_sr = trainer.processor.feature_extractor.sampling_rate
 
- # compute log-Mel input features from input audio array
- batch["input_features"] = trainer.feature_extractor(
- audio["array"], sampling_rate=audio["sampling_rate"]
- ).input_features[0]
+ def prepare_dataset(batch):
+ # load and resample audio data from 48 to 16kHz
+ audio, _ = librosa.load(batch["audio"], sr=target_sr)
 
- # encode target text to label ids
- batch["labels"] = trainer.tokenizer(batch["sentence"]).input_ids
- return batch
+ # compute log-Mel input features from input audio array
+ batch["input_features"] = trainer.feature_extractor(
+ audio, sampling_rate=target_sr
+ ).input_features[0]
 
+ # encode target text to label ids
+ batch["labels"] = trainer.tokenizer(batch["lyrics"]).input_ids
+ return batch
 
-common_voice = common_voice.map(
- prepare_dataset, remove_columns=common_voice.column_names["train"], num_proc=1
-)
+ dataset = dataset.map(prepare_dataset, num_proc=1)
 
-trainer.train(common_voice)
+ # save the processed dataset
+ dataset.save_to_disk("dataset/test/")
+
+else:
+ # load the processed dataset
+ dataset = load_dataset("dataset/test/")
+
+dataset = dataset.train_test_split(test_size=0.05)
+trainer.train(dataset)

training/train.py

@@ -71,7 +71,7 @@ def _compute_metrics(self, pred):
  label_str = self.tokenizer.batch_decode(label_ids, skip_special_tokens=True)
 
  wer = 100 * self.metric.compute(predictions=pred_str, references=label_str)
-
+ print(f"WER: {wer}")
  return {"wer": wer}
 
  def train(self, dataset):
@@ -93,8 +93,8 @@ def train(self, dataset):
  per_device_eval_batch_size=8,
  predict_with_generate=True,
  generation_max_length=225,
- save_steps=10,
- eval_steps=10,
+ save_steps=80,
+ eval_steps=40,
  logging_steps=25,
  report_to=["tensorboard"],
  load_best_model_at_end=True,