Update scripts/prep_push_to_hf.py

scripts/prep_push_to_hf.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-import os
+from os.path import join
 from pathlib import Path
 
 import huggingface_hub
@@ -8,38 +8,38 @@ from datasets import Audio, Dataset
 import pandas as pd
 
 MAX_DURATION_IN_SECONDS = 10.0
+MAX_LEN = 500
 
 
 def duration_filter(item):
     return item < MAX_DURATION_IN_SECONDS
 
 
+def text_filter(item):
+    return len(item) < MAX_LEN
+
+
 repo_dir = Path(__file__).resolve().parent.parent
-data_dir = os.path.join(repo_dir, "data")
-kendex_dir = os.path.join(data_dir, "Kendex")
-audio_dir = os.path.join(kendex_dir, "wavs")
-metadata = os.path.join(kendex_dir, "metadata.csv")
-
-df = pd.read_csv(metadata, delimiter="|", header=None)
-audio = pd.Series([os.path.join(audio_dir, f"{f}.wav") for f in df[0]])
-
-dataset = Dataset.from_pandas(
-    pd.DataFrame(
-        {
-            "audio": audio,
-            "text": df[1],
-            "duration": [librosa.get_duration(path=a) for a in audio],
-        }
-    )
+data_dir = join(repo_dir, "data")
+kendex_dir = join(data_dir, "Kendex")
+audio_dir = join(kendex_dir, "wavs")
+
+metadata = pd.read_csv(join(kendex_dir, "metadata.csv"), delimiter="|", header=None)
+wavs = pd.Series([join(audio_dir, f"{f}.wav") for f in metadata[0]])
+
+df = pd.DataFrame(
+    {
+        "audio": wavs,
+        "file": wavs,
+        "text": wavs[1],
+        "duration": [librosa.get_duration(path=a) for a in wavs],
+    }
 )
 
+dataset = Dataset.from_pandas(df)
 dataset = dataset.cast_column("audio", Audio(sampling_rate=16_000))
-
-dataset = dataset.filter(
-    lambda x: False if len(x) >= 500 else True, input_columns=["text"]
-)
+dataset = dataset.filter(text_filter, input_columns=["text"])
 dataset = dataset.filter(duration_filter, input_columns=["duration"])
-
 dataset = dataset.remove_columns(["duration"])
 
 huggingface_hub.login()