ovos-audio-transformer-plugin-bandpass
September 1, 2026 · View on GitHub
A band-pass filter audio transformer for OpenVoiceOS. It runs on the captured speech audio before STT and attenuates energy outside a configurable pass-band.
The default 300–3400 Hz band is the telephone speech band: it removes low-frequency rumble (HVAC, handling noise) and high-frequency hiss that carry no phonetic information, which can improve recognition on noisy inputs without touching the speech formants.
Install
pip install ovos-audio-transformer-plugin-bandpass
Configure
Enable it in mycroft.conf under audio_transformers. Any component that runs
the audio-transformer pipeline, such as the listener or the STT server, applies it:
{
"audio_transformers": {
"ovos-audio-transformer-plugin-bandpass": {
"low_hz": 300,
"high_hz": 3400,
"order": 4,
"sample_rate": 16000
}
}
}
| Option | Default | Meaning |
|---|---|---|
low_hz | 300 | Lower edge of the pass-band, in Hz. |
high_hz | 3400 | Upper edge of the pass-band, in Hz. |
order | 4 | Butterworth filter order. Higher is a steeper roll-off. |
sample_rate | 16000 | Sample rate of the incoming PCM audio, in Hz. |
Edges are clamped into the open interval below Nyquist, so an aggressive band on a low sample rate degrades gracefully instead of failing.
Related projects
- OpenVoiceOS/ovos-plugin-manager: loads audio transformer plugins through the
opm.transformer.audioentry point. - OpenVoiceOS/ovos-audio-transformer-plugin-speechbrain-langdetect: another audio transformer plugin, for language detection.
License
Apache-2.0. See LICENSE.