Descript Audiotools

0.7.2 · active · verified Wed Apr 15

Descript Audiotools is a Python library providing object-oriented handling of audio data, featuring functionalities like fast augmentation routines, batching, padding, and GPU-powered operations. The current stable version available on PyPI is 0.7.2. The library is actively maintained, with development continuing on its GitHub repository.

Warnings

breaking Version 0.7.2 of `descript-audiotools` has a strict dependency on `protobuf<3.20,>=3.9.2`. Installing with newer `protobuf` versions (e.g., 4.x or 5.x) will lead to dependency conflicts and prevent installation or cause runtime errors.
Fix: Ensure your environment uses `protobuf` within the `3.9.2` to `3.19.x` range. You may need to downgrade `protobuf` if other packages require a newer version, or consider installing `audiotools` directly from GitHub if a newer, compatible version exists there (e.g., `pip install git+https://github.com/descriptinc/audiotools@main`).
gotcha The `AudioSignal.play()` method relies on the external `ffplay` command-line utility for audio playback. If `ffplay` (part of FFmpeg) is not installed on your system or not in your system's PATH, `signal.play()` calls will fail silently or with an error about the command not being found.
Fix: Install FFmpeg (which includes `ffplay`) on your operating system and ensure its executables are accessible via your system's PATH. Common installation methods include `brew install ffmpeg` on macOS, `sudo apt-get install ffmpeg` on Debian/Ubuntu, or downloading from the official FFmpeg website for Windows.

Install

pip install descript-audiotools Install from PyPI

Imports

AudioSignal
wrong:
```
import audiotools.AudioSignal
```
correct:
```
from audiotools import AudioSignal
```
AudioSignal is a class directly exported from the top-level package.
audiotools
```
import audiotools
```

Quickstart

This quickstart demonstrates how to create a basic `AudioSignal` object, apply a simple audio transformation (low-pass filter), and print its properties. For actual audio playback using `signal.play()`, the `ffplay` utility (part of the FFmpeg suite) needs to be installed and accessible in your system's PATH. The example generates a dummy sine wave for immediate execution without external audio files.

import audiotools
import numpy as np

# Create a dummy AudioSignal (e.g., a sine wave)
sample_rate = 44100
duration = 3 # seconds
frequency = 440 # Hz
t = np.linspace(0, duration, int(sample_rate * duration), endpoint=False)
sine_wave = 0.5 * np.sin(2 * np.pi * frequency * t)

# AudioSignal expects a torch.Tensor, so convert numpy array
signal = audiotools.AudioSignal(sine_wave[np.newaxis, :], sample_rate)

print(f"Original signal sample rate: {signal.sample_rate}")
print(f"Original signal duration: {signal.duration} seconds")

# Apply a low-pass filter
filtered_signal = signal.low_pass(8000)
print(f"Filtered signal duration: {filtered_signal.duration} seconds")

# To play back audio, ffplay (part of FFmpeg) must be installed.
# signal.play() 
# filtered_signal.play()

view raw JSON →