Skip to content

PhysioKit

Package version Supported Python versions Package downloads Package downloads License

Documentation: https://ambiqai.github.io/physiokit

Source Code: https://github.com/AmbiqAI/physiokit

Key Features

  • Multi-signal coverage: ECG, PPG, RSP, IMU, and HRV.
  • Wearable-first: robust to noisy, real-time data.
  • Batteries included: cleaning, peak detection, metrics, and synthesis.
  • Extensible utilities: shared filters, smoothing, FFT, and noise/distortion.

PhysioKit is built to be modular: each signal family has a consistent API for cleaning, peak finding, intervals, and downstream metrics. Shared pk.signal primitives keep filtering and resampling uniform so you can mix modalities without rewriting glue code. Synthetic generators let you prototype quickly or build regression data for your own pipelines.

Requirements

Installation

Installing PhysioKit can be done using uv or pip.

$ uv add physiokit

---> 100%


$ pip install physiokit

---> 100%

Quickstart

import numpy as np
import physiokit as pk

fs = 500
ecg, segs, fids = pk.ecg.synthesize(signal_length=5*fs, sample_rate=fs, heart_rate=70, leads=1)
ecg = ecg.squeeze()

ecg_clean = pk.ecg.clean(ecg, sample_rate=fs)
peaks = pk.ecg.find_peaks(ecg_clean, sample_rate=fs)
rri = pk.ecg.compute_rr_intervals(peaks)
mask = pk.ecg.filter_rr_intervals(rri, sample_rate=fs)

hr_bpm, _ = pk.ecg.compute_heart_rate(ecg_clean, sample_rate=fs)
hrv_td = pk.hrv.compute_hrv_time(rri[mask == 0], sample_rate=fs)

import numpy as np
import physiokit as pk

fs = 100
t = np.arange(0, 10, 1/fs)
ppg = np.sin(2*np.pi*1.1*t)

hr_bpm, qos = pk.ppg.compute_heart_rate(ppg, sample_rate=fs, method="peak")
spo2 = pk.ppg.compute_spo2_in_time(ppg, ppg, sample_rate=fs)

import numpy as np
import physiokit as pk

fs = 25
t = np.arange(0, 40, 1/fs)
rsp = np.sin(2*np.pi*0.2*t)

bpm, qos = pk.rsp.compute_respiratory_rate(rsp, sample_rate=fs, method="peak")

import numpy as np
import physiokit as pk

fs = 100
t = np.arange(0, 5, 1/fs)
sig = np.sin(2*np.pi*2*t) + 0.2*np.random.randn(t.size)

clean = pk.signal.filter_signal(sig, lowcut=0.5, highcut=20, sample_rate=fs)
freqs, sp = pk.signal.compute_fft(clean, sample_rate=fs)

License

This project is licensed under the terms of BSD 3-Clause.