Gammatone Filterbank

1.0.3 verified Fri May 01 auth: no python

A Python implementation of gammatone filters for auditory modeling, including filterbank generation and cochleagram computation. Current version: 1.0.3 (patch release, adds f_max parameter to gtgram). Requires Python >=3.8.

pip install gammatone

Common errors

error ImportError: cannot import name 'Gammatone' from 'gammatone' ↓

cause Trying to import Gammatone from the top-level package instead of the implementation module.

fix

Use 'from gammatone.gammatone_impl import Gammatone'

error ValueError: The number of channels is too high for the given f_max ↓

cause Requesting more channels than possible given the frequency range when using gtgram.

fix

Reduce channels or increase f_max. Ensure f_max / f_min >= 2^(channels/ERB scale).

error TypeError: gtgram() got an unexpected keyword argument 'f_max' ↓

cause Using an older version of gammatone (<1.0.3) that does not support the f_max parameter.

fix

Upgrade: pip install --upgrade gammatone

Warnings

gotcha The default f_max in gtgram is 4000 Hz. If you need higher frequencies (e.g., for 22050 Hz audio), pass f_max explicitly. Added in v1.0.3. ↓

fix Call gtgram(... , f_max=8000) to extend coverage.

gotcha Input signal must be 1-D array. If you pass a 2-D array (e.g., stereo audio), it will fail silently or produce incorrect shape. ↓

fix Mono mix: signal = np.mean(signal, axis=1) for stereo.

deprecated Direct import from gammatone.gammatone (without _impl suffix) may be deprecated in future releases. Use gammatone.gammatone_impl for core filter classes. ↓

fix Use from gammatone.gammatone_impl import Gammatone

Imports

Gammatone

from gammatone.gammatone_impl import Gammatone

gtgram
```
from gammatone.gtgram import gtgram
```

gammatone_freq

from gammatone.filters import make_erb_filters, centre_freqs

Quickstart

Create a gammatone-based cochleagram from an audio signal.

import numpy as np
from gammatone.gtgram import gtgram

# Generate a cochleagram from a sample audio signal
fs = 16000
duration = 0.1
t = np.linspace(0, duration, int(fs*duration), endpoint=False)
signal = np.sin(2*np.pi*1000*t)  # 1 kHz tone

# Compute gammatone spectrogram (cochleagram)
coch = gtgram(signal, fs, window_time=0.02, hop_time=0.01, channels=64, f_min=50, f_max=8000)
print(coch.shape)