can't read annotations

Hi everybody. I have annotations for EEG signal in .txt file. The signals annotate every 30-s with a sampling rate of 250. I extracted sample and Aux columns. Then I extracted the starting and ending sample of each event (‘H’, 'HA) to segment the signal. I used pandas to do that and then converted the dataframe to .txt file. Now I’m trying to read it by mne.read_annotations which can read .txt files, but it doesn’t work. What is the problem? Is there any other way to find events and epoch the signal?
I hope that I provide all the required informations. If not, please tell me to add more. Thank you.

By the way, I’m working on MIT-BIH Polysomnographic Database. If someone worked on it before, I’d appreciate it if help me to preprocess the signal.

  • MNE version: 0.24.1
  • operating system: Google Colab

You might want to check out SleepECG – we’re currently working on providing readers for various sleep database records, and it could be that the record you want to load is already implemented.

If this does not work, then the format of your .txt file is probably not what mne.read_annotations expects. You’d have to provide more details on how your file is structured, and what the exact error message is.

Thank you Clemens. I checked the link but I didn’t find the records I was looking for. The data I’m working on is: slpdb

The .txt file I extracted is like this:

index    Sample #      Aux                  
 12        90000     3 HA
 13        97500   3 H LA
 14       105000         
 36       270000  4 LA HA
 37       277500         
 46       345000      2 H
 47       352500         
 49       367500      2 H
 50       375000         
 163     1222500     2 HA
 164     1230000         
 167     1252500   2 H LA
 168     1260000         
 174     1305000      2 H
 175     1312500         
 179     1342500     2 HA
 180     1350000  2 LA HA
 181     1357500   2 H LA
 182     1365000         
 185     1387500   2 LA H
 186     1395000         
 187     1402500      2 H
 188     1410000      3 H
 189     1417500     3 HA
 ...       ...        ...

And the error is:

ValueError: Invalid value for the ‘text header’ parameter. Allowed values are 3, and 4, but got 58 instead.

Actually, there is, which I think reads the data you want.

1 Like

Can I use it in Google Colab? I tried to use it but I couldn’t.


The wfdb package allows to read record and annotation files from PhysioNet. This minimal example shows how you could access annotations and annotation times. It requires the .hea,.dat and .st files for the desired record.

import wfdb
import mne

# change the path if necessary
# note that this path does _not_ include any file extension, as it's added automatically by `wfdb.rdrecord`
record_path = '../../datasets/slpdb/slp01a'

record = wfdb.rdrecord(record_path)
annotation = wfdb.rdann(record_path, 'st')

# annotation.sample contains the annotation's sample indices
# annotation.aux_note contains sleep stage, apnea and movement information
# slpdb doesn't provide annotation durations, so let's assume 30 seconds
annotation_mne = mne.Annotations(
    onset=annotation.sample / record.fs,

I read the annotations successfully. Thank you so much Florian!