add command to validate a BIDS folder has all the extracted features

src/b2aiprep/cli.py

@@ -32,7 +32,7 @@
 from streamlit.web.bootstrap import run
 
 from b2aiprep.prepare.bids_like_data import redcap_to_bids
-from b2aiprep.prepare.prepare import prepare_bids_like_data
+from b2aiprep.prepare.prepare import prepare_bids_like_data, validate_bids_data
 
 
 @click.group()
@@ -119,6 +119,28 @@ def prepbidslikedata(
     )
 
 
+@main.command()
+@click.argument("bids_dir_path", type=click.Path())
+@click.argument("fix", type=bool)
+def validate(
+    bids_dir_path,
+    fix,
+):
+    """Organizes the data into a BIDS-like directory structure.
+
+    redcap_csv_path: path to the redcap csv\n
+    audio_dir_path: path to directory with audio files\n
+    bids_dir_path: path to store bids-like data\n
+    tar_file_path: path to store tar file\n
+    transcription_model_size: tiny, small, medium, or large\n
+    n_cores: number of cores to run feature extraction on\n
+    with_sensitive: whether to include sensitive data
+    """
+    validate_bids_data(
+        bids_dir_path=Path(bids_dir_path),
+        fix=fix,
+    )
+
 @main.command()
 @click.argument("filename", type=click.Path(exists=True))
 @click.option("-s", "--subject", type=str, default=None)

src/b2aiprep/prepare/prepare.py

@@ -348,6 +348,80 @@ def prepare_bids_like_data(
 
     _logger.info("Process completed.")
 
+def validate_bids_data(
+    bids_dir_path: Path,
+    fix: bool = True,
+    transcription_model_size: str = 'medium',
+) -> None:
+    """Scans BIDS audio data and verifies that all expected features are present."""
+    _logger.info("Scanning for features in BIDS directory.")
+    # TODO: add a check to see if the audio feature extraction is complete
+    # before proceeding to the next step
+    # can verify features are generated for each audio_dir by looking for .pt files
+    # in audio_dir.parent / "audio"
+    audio_paths = get_audio_paths(bids_dir_path)
+    audio_to_reprocess = defaultdict(list)
+    features = ('speaker_embedding', 'specgram', 'melfilterbank', 'mfcc', 'sample_rate', 'opensmile')
+    for audio_path in audio_paths:
+        audio_path = Path(audio_path)
+        audio_dir = audio_path.parent
+        features_dir = audio_dir.parent / "audio"
+        for feat_name in features:
+            if features_dir.joinpath(f'{audio_path.stem}_{feat_name}.pt').exists() is False:
+                audio_to_reprocess[audio_path].append(feat_name)
+
+        # also check for transcription
+        if features_dir.joinpath(f'{audio_path.stem}_transcription.txt').exists() is False:
+            audio_to_reprocess[audio_path].append('transcription')
+
+    if len(audio_to_reprocess) > 0:
+        _logger.info(f"Missing features for {len(audio_to_reprocess)} / {len(audio_paths)} audio files")
+    else:
+        _logger.info("All audio files have been processed and all feature files are present.")
+        return
+
+    if not fix:
+        return
+
+    feature_extraction_fcns = {
+        "speaker_embedding": extract_speaker_embeddings_from_audios,
+        "specgram": extract_spectrogram_from_audios,
+        "melfilterbank": extract_mel_filter_bank_from_audios,
+        "mfcc": extract_mfcc_from_audios,
+        "opensmile": extract_opensmile_features_from_audios,
+    }
+    for audio_path, missing_feats in tqdm(audio_to_reprocess.items(), total=len(audio_to_reprocess), desc='Reprocessing audio files'):
+        audio_dir = audio_path.parent
+        features_dir = audio_dir.parent / "audio"
+        features_dir.mkdir(exist_ok=True)
+        for feat_name in missing_feats:
+            audio = Audio.from_filepath(str(audio_path))
+            audio = resample_audios([audio], resample_rate=RESAMPLE_RATE)[0]
+
+            file_extension = "pt"
+            if feat_name == "sample_rate":
+                feature_value = audio.sampling_rate
+            elif feat_name in feature_extraction_fcns:
+                feature_value = feature_extraction_fcns[feat_name]([audio])[0]
+            elif feat_name == 'transcription':
+                language = Language.model_validate({"language_code": "en"})
+                speech_to_text_model = HFModel(path_or_uri=f"openai/whisper-{transcription_model_size}")
+                feature_value = transcribe_audios(
+                    audios=[audio], model=speech_to_text_model, language=language
+                )[0]
+                file_extension = "txt"
+            else:
+                _logger.warning(f"Unsupported feature: {feat_name}")
+                continue
+
+            save_path = features_dir / f"{audio_path.stem}_{feat_name}.{file_extension}"
+            if file_extension == "pt":
+                torch.save(feature_value, save_path)
+            else:
+                with open(save_path, "w", encoding="utf-8") as text_file:
+                    text_file.write(feature_value.text)
+
+    _logger.info("Process completed.")
 
 def main():
     pass