High-quality Audio Sample Rate Conversion

Warnings

• gotcha When installing on Linux, the underlying `libsamplerate` C library may need to be installed separately (e.g., `libsamplerate0` on Debian/Ubuntu, `libsamplerate` on Arch Linux). Binary wheels for macOS and Windows include `libsamplerate` statically built.
  Fix: Ensure `libsamplerate` is installed via your system's package manager on Linux: `sudo apt-get install libsamplerate0` (Debian/Ubuntu) or `sudo pacman -S libsamplerate` (Arch).
• gotcha The `input_data` must be of `float32` dtype and C-contiguous for optimal performance and correct interaction with the underlying C library.
  Fix: Convert your input NumPy array to `np.float32` using `array.astype(np.float32)` and ensure it's C-contiguous.
• gotcha There is an older, distinct package named `scikits.samplerate` which also provides audio resampling. While the `resample` function signature is compatible, `scikits.samplerate` only implements the Simple API and uses Cython, whereas `samplerate` (this package) uses `pybind11` and exposes all three libsamplerate APIs.
  Fix: Ensure you are installing `samplerate` (the package described here) and not `scikits.samplerate` if you intend to use the Full or Callback APIs or benefit from the `pybind11` implementation.

Install

• pip install samplerate Install with pip

Imports

• samplerate

  import samplerate

Quickstart

This example demonstrates both the Simple API for single-call resampling and the Full API for chunk-based processing, using a synthesized sine wave. It also includes an assertion to show that both APIs yield the same result for a complete signal.

import numpy as np
import samplerate

# Synthesize data
fs = 1000.0
t = np.arange(fs * 2) / fs
input_data = np.sin(2 * np.pi * 5 * t).astype(np.float32)

# Simple API
ratio = 1.5
converter = 'sinc_best'
output_data_simple = samplerate.resample(input_data, ratio, converter)

print(f"Original shape: {input_data.shape}, Resampled shape (Simple API): {output_data_simple.shape}")

# Full API
resampler = samplerate.Resampler(converter, channels=1)
output_data_full = resampler.process(input_data, ratio, end_of_input=True)

print(f"Resampled shape (Full API): {output_data_full.shape}")

assert np.allclose(output_data_simple, output_data_full)