Pandas Audio Methods

Audio methods for pandas dataframes using soundfile.

Features:

• Use sf.SoundFile objects in pandas dataframes
• Call sf.SoundFile methods on a column, for example:
  • .read()
  • .truncate()
  • .seek()
• Save dataframes with sf.SoundFile objects to Parquet
• Process Audios in parallel with Dask
• Manipulate Audio datasets from Hugging Face

Installation

pip install pandas-audio-methods

Usage

You can open Audios as sf.SoundFile objects using the .open() method.

Once the Audios are opened, you can call any sf.SoundFile method:

import pandas as pd
from pandas_audio_methods import SFMethods

pd.api.extensions.register_series_accessor("sf")(SFMethods)

df = pd.DataFrame({"file_path": ["path/to/audio.wav"]})
df["audio"] = df["file_path"].sf.open()
# 0    SoundFile('path/to/audio.wav', mode='r', sampl...
# Name: audio, dtype: object, soundfile methods enabled

Use with librosa:

import librosa
df["audio"] = [librosa.load(audio, sr=16_000) for audio in df["audio"]]
df["audio"] = df["audio"].sf.write()
# 0    SoundFile(<_io.BytesIO object at 0x11b747ba0>,...
# Name: audio, dtype: object, soundfile methods enabled

Here is how to enable sf methods for sf.SoundFiles created manually:

df = pd.DataFrame({"audio": [sf.SoundFile.open("path/to/audio.wav")]})
df["audio"] = df["audio"].sf.enable()
# 0    SoundFile('path/to/audio.wav', mode='r', sampl...
# Name: Audio, dtype: object, soundfile methods enabled

Save

You can save a dataset of sf.SoundFiles to Parquet:

# Save
df = pd.DataFrame({"file_path": ["path/to/audio.wav"]})
df["audio"] = df["file_path"].sf.open()
df.to_parquet("data.parquet")

# Later
df = pd.read_parquet("data.parquet")
df["audio"] = df["audio"].sf.enable()

This doesn't just save the paths to the Audio files, but the actual Audios themselves !

Under the hood it saves dictionaries of `{"bytes": , "path":