Simple wrapper for librubberband
Project description
Introduction
rubberband is a simple Python3 wrapper around the well-known librubberband sound stretching / pitch-shifting library. Unlike existing Python wrappers (e.g. pyrubberband) this is a true native extension.
The initial release provides a single function that will stretch a mono audio stream by multiplying its duration by a provided factor. Future versions may include pitch shifting, and more complex processing based on data maps.
Installation
The module is available only for MacOS and Linux. The code may compile on Windows, but it has not been tested. Dependencies are:
Python 3 (preferably 3.6 or greater)
librubberband (> 1.8)
libsndfile (> 1.0)
Assuming these dependencies are met, then installation is simplicity itself:
pip3 install rubberband
The install script does check for the required libraries, and will complain vociferously if they cannot be located. Information on obtaining them is available from the links above.
API
Audio stream formatting
The module exposes a single function rubberband.stretch which applies the librubberband algorithm to a mono audio stream, encoded in one of the following formats, each of which has a corresponding constant, as set out in the third column:
Format |
Description |
Constant |
---|---|---|
PCM_U8 |
unsigned 8-bit |
rubberband.uint8 |
PCM_S8 |
signed 8-bit |
rubberband.int8 |
PCM_16 |
signed 16-bit |
rubberband.int16 |
PCM_32 |
signed 32-bit |
rubberband.int32 |
FLOAT |
normalised 32-bit float |
rubberband.float32 |
Note that floating point data is assumed to be normalised, so all samples lie in the range [-1,1).
Audio data can be passed to rubberband.stretch in any of three ways:
- Typed array
A 1-dimensional NUMPY typed array object, whose dtype is one of numpy.uint8, numpy.int8, numpy.int16, numpy.int32 or numpy.float32. The type of the audio data is deduced from this, using the strangely convenient fact that if T is one of uint8, int8, int16, int32, float32 then
numpy.dtype(numpy.T).num == rubberband.T
- List
A simple Python list, all of whose elements are of a type implicitly convertible to float. In this case, the audio format cannot be deduced, so it must be specified using the format argument to rubberband.stretch (see below).
- Raw bytestream
A Python bytes object, whose content is the raw PCM byte stream (note: audio file metadata, e.g. WAV file headers, must be stripped, so only PCM data remains). Again, in this case, the audio format cannot be deduced, so it must be specified using the format argument to rubberband.stretch (see below).
In all cases, the output from rubberband.stretch has the same PCM format, and is stored in the same kind of object, as the input. So, for example, given a bytes object representing a PCM_16 audio channel, rubberband.stretch returns bytes object representing a PCM_16 stretched audio channel.
Method signature
rubberband.stretch (input, format = rubberband.float32, rate = 48000 , ratio = 1 , crispness = 5 , formants = False, precise = True )
Arguments
- input
The input is assumed to represent a single channel of PCM audio data, encoded with one of the schemes listed above. It can be any of the types set out above.
- format
The PCM format of the data, specified using one of the constants set out above. This value is ignored if input is a NUMPY typed array, in which case the format is deduced from its dtype.
- rate
The frame rate of the input audio stream (so bit rate divided by sample size).
- ratio
The ratio of output length to input length (in seconds / number of samples).
- crispness
Integer 0 - 6, default 5: measure of performance - see the rubberband-cli documentation for more details.
- formants
Boolean, default False : whether or not to preserve formants - see the rubberband-cli documentation for more details.
- precise
Boolean, default True : whether or not to use the precise stretching algorithm - see the rubberband-cli documentation for more details.
- Return value
An object containing the stretched audio data, represented using the same PCM encoding as the input. Samples are normalised to lie in the expected range for the format.
Example
import rubberband import soundfile data, rate = soundfile.read('infile.wav',dtype='int16') bitrate=rate*16 nFrames=len(data) oldDuration=nFrames/rate newDuration=6 ratio=newDuration/oldDuration print(f'Raw input type is : {type(stream)}') out=rubberband.stretch(stream,rate=rate,ratio=ratio,crispness=5, formants=False,precise=True) soundfile.write('outfile.wav',out,rate,'PCM_16')
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.