[Examples] Use Audio feature in speech classification (#14052)

* Update SEW integration test tolerance

* Update audio classification

* Update test

* Remove torchaudio

* Add dataset revision

* Hub branch naming

* Revert dataset revisions

* Update datasets

examples/pytorch/audio-classification/README.md

@@ -68,7 +68,7 @@ The following command shows how to fine-tune [wav2vec2-base](https://huggingface
 ```bash
 python run_audio_classification.py \
     --model_name_or_path facebook/wav2vec2-base \
-    --dataset_name anton-l/common_language \
+    --dataset_name common_language \
     --audio_column_name path \
     --label_column_name language \
     --output_dir wav2vec2-base-lang-id \

examples/pytorch/audio-classification/requirements.txt

-datasets>=1.12.0
+datasets>=1.14.0
+librosa
 torchaudio
 torch>=1.6
\ No newline at end of file

examples/pytorch/audio-classification/run_audio_classification.py

@@ -22,7 +22,6 @@ from typing import Optional
 
 import datasets
 import numpy as np
-import torchaudio
 from datasets import DatasetDict, load_dataset
 
 import transformers
@@ -43,19 +42,9 @@ from transformers.utils.versions import require_version
 logger = logging.getLogger(__name__)
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
-check_min_version("4.11.0.dev0")
+check_min_version("4.12.0.dev0")
 
-require_version("datasets>=1.12.1", "To fix: pip install -r examples/pytorch/audio-classification/requirements.txt")
-
-
-def load_audio(path: str, sample_rate: int = 16000):
-    wav, sr = torchaudio.load(path)
-    # convert multi-channel audio to mono
-    wav = wav.mean(0)
-    # standardize sample rate if it varies in the dataset
-    resampler = torchaudio.transforms.Resample(sr, sample_rate)
-    wav = resampler(wav)
-    return wav
+require_version("datasets>=1.14.0", "To fix: pip install -r examples/pytorch/audio-classification/requirements.txt")
 
 
 def random_subsample(wav: np.ndarray, max_length: float, sample_rate: int = 16000):
@@ -100,8 +89,8 @@ class DataTrainingArguments:
         },
     )
     audio_column_name: Optional[str] = field(
-        default="file",
-        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'file'"},
+        default="audio",
+        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
     )
     label_column_name: Optional[str] = field(
         default="label", metadata={"help": "The name of the dataset column containing the labels. Defaults to 'label'"}
@@ -246,13 +235,18 @@ def main():
         use_auth_token=True if model_args.use_auth_token else None,
     )
 
+    # `datasets` takes care of automatically loading and resampling the audio,
+    # so we just need to set the correct target sampling rate.
+    raw_datasets = raw_datasets.cast_column(
+        data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+    )
+
     def train_transforms(batch):
         """Apply train_transforms across a batch."""
         output_batch = {"input_values": []}
-        for f in batch[data_args.audio_column_name]:
-            wav = load_audio(f, sample_rate=feature_extractor.sampling_rate)
+        for audio in batch[data_args.audio_column_name]:
             wav = random_subsample(
-                wav, max_length=data_args.max_length_seconds, sample_rate=feature_extractor.sampling_rate
+                audio["array"], max_length=data_args.max_length_seconds, sample_rate=feature_extractor.sampling_rate
             )
             output_batch["input_values"].append(wav)
         output_batch["labels"] = [label for label in batch[data_args.label_column_name]]
@@ -262,8 +256,8 @@ def main():
     def val_transforms(batch):
         """Apply val_transforms across a batch."""
         output_batch = {"input_values": []}
-        for f in batch[data_args.audio_column_name]:
-            wav = load_audio(f, sample_rate=feature_extractor.sampling_rate)
+        for audio in batch[data_args.audio_column_name]:
+            wav = audio["array"]
             output_batch["input_values"].append(wav)
         output_batch["labels"] = [label for label in batch[data_args.label_column_name]]
 
@@ -311,8 +305,6 @@ def main():
         model.freeze_feature_extractor()
 
     if training_args.do_train:
-        if "train" not in raw_datasets:
-            raise ValueError("--do_train requires a train dataset")
         if data_args.max_train_samples is not None:
             raw_datasets["train"] = (
                 raw_datasets["train"].shuffle(seed=training_args.seed).select(range(data_args.max_train_samples))
@@ -321,8 +313,6 @@ def main():
         raw_datasets["train"].set_transform(train_transforms, output_all_columns=False)
 
     if training_args.do_eval:
-        if "eval" not in raw_datasets:
-            raise ValueError("--do_eval requires a validation dataset")
         if data_args.max_eval_samples is not None:
             raw_datasets["eval"] = (
                 raw_datasets["eval"].shuffle(seed=training_args.seed).select(range(data_args.max_eval_samples))

examples/pytorch/speech-pretraining/run_wav2vec2_pretraining_no_trainer.py

@@ -113,7 +113,7 @@ def parse_args():
         "--audio_column_name",
         type=str,
         default="audio",
-        help="Column in the dataset that contains speech file path. Defaults to 'file'",
+        help="Column in the dataset that contains speech file path. Defaults to 'audio'",
     )
     parser.add_argument(
         "--model_name_or_path",
@@ -431,9 +431,9 @@ def main():
     # via the `feature_extractor`
     feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(args.model_name_or_path)
 
-    # make sure that dataset decodes audio with correct samlping rate
+    # make sure that dataset decodes audio with correct sampling rate
     raw_datasets = raw_datasets.cast_column(
-        "audio", datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
     )
 
     # only normalized-inputs-training is supported

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -454,9 +454,9 @@ def main():
     # so that we just need to set the correct target sampling rate and normalize the input
     # via the `feature_extractor`
 
-    # make sure that dataset decodes audio with correct samlping rate
+    # make sure that dataset decodes audio with correct sampling rate
     raw_datasets = raw_datasets.cast_column(
-        "audio", datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
     )
 
     # derive max & min input length for sample rate & max duration

examples/pytorch/test_examples.py

@@ -428,7 +428,7 @@ class ExamplesTests(TestCasePlus):
             --dataset_config_name ks
             --train_split_name test
             --eval_split_name test
-            --audio_column_name file
+            --audio_column_name audio
             --label_column_name label
             --do_train
             --do_eval