Cheatsheet: Loading the PARSEL Corpus

MDIG2026

Setup

Python

import os, re, glob
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

CORPUS    = "../.."
acoustics = f"{CORPUS}/TS_acoustics"
meta_path = f"{CORPUS}/data/Intake_Survey/cleaned_curated_Intake_Session_new.csv"
data_path = f"{CORPUS}/data/Conversations"

def extract_pid(filename):
    match = re.search(r"PID_\d+", filename)
    return match.group() if match else None
  
def mask_from_spans(spans, time):
    m = np.zeros(len(time), dtype=bool)
    for on, off in spans[["onset", "offset"]].itertuples(index=False):
        m |= (time >= on) & (time <= off)
    return m

R

library(tidyverse)

CORPUS    <- "../.."


acoustics <- file.path(CORPUS, "TS_acoustics")

Metadata

meta = pd.read_csv(meta_path)
meta.head()

   Unnamed: 0  Age_part Sex_part  ... Accuracy_total globalPID UCMRT
0           0        35   Female  ...              5    PID_58   0.0
1           1        41     Male  ...              6    PID_59   6.0
2           2        67   Female  ...              6    PID_43   9.0
3           3        38   Female  ...              6    PID_42  20.0
4           4        64     Male  ...              6    PID_26  12.0

[5 rows x 172 columns]

Acoustics

Amplitude envelope

time is in milliseconds.

envelope_files = glob.glob(f"{acoustics}/env_*.csv")
pcn = extract_pid(os.path.basename(envelope_files[0]))
env = pd.read_csv(envelope_files[0])
env["time_s"] = env["time"] / 1000.0
print(env.shape)

(90660, 5)

env.head()

   time     audio  ...                                           filename time_s
0   0.0  0.000000  ...  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-Batc...  0.000
1   2.0  0.000000  ...  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-Batc...  0.002
2   4.0  0.000000  ...  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-Batc...  0.004
3   6.0  0.000000  ...  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-Batc...  0.006
4   8.0  0.000031  ...  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-Batc...  0.008

[5 rows x 5 columns]

fig, ax = plt.subplots(figsize=(10, 2.4))
ax.plot(env["time_s"], env["envelope"], lw=0.8)
ax.set_xlabel("time (s)"); ax.set_ylabel("envelope")
plt.tight_layout(); plt.show()

envelope_files <- list.files(acoustics, pattern = "^env_.*\\.csv$", full.names = TRUE)
env <- read_csv(envelope_files[1]) |> mutate(time_s = time / 1000)
head(env)

# A tibble: 6 × 5
   time     audio   envelope filename                                     time_s
  <dbl>     <dbl>      <dbl> <chr>                                         <dbl>
1     0 0         0.00000747 conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0    
2     2 0         0.00000829 conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0.002
3     4 0         0.00000913 conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0.004
4     6 0         0.00000998 conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0.006
5     8 0.0000305 0.0000108  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0.008
6    10 0.0000305 0.0000116  conv-self-LPID_1-Batch_10-LPID_2-Batch_10-B…  0.01 

F0 and voicing on/offsets

f0 is blank where unvoiced. Rising/falling edges of that give voiced segments.

def voicing_segments(f0_path):
    df = pd.read_csv(f0_path)
    t = df["time_ms"].to_numpy() / 1000.0
    voiced = df["f0"].notna().to_numpy()
    edges = np.diff(voiced.astype(int))
    onsets, offsets = t[1:][edges == 1], t[1:][edges == -1]
    if voiced[0]:  onsets  = np.r_[t[0], onsets]
    if voiced[-1]: offsets = np.r_[offsets, t[-1]]
    return pd.DataFrame({"onset": onsets, "offset": offsets})

voicing = []
for f in glob.glob(f"{acoustics}/f0_*.csv"):
    p = extract_pid(os.path.basename(f))
    seg = voicing_segments(f)
    seg["pcn"] = p
    voicing.append(seg)
voicing = pd.concat(voicing, ignore_index=True)

print(f"{voicing.groupby(['pcn']).ngroups} trials, {len(voicing)} voiced segments")

6 trials, 33990 voiced segments

voicing.head()

      onset    offset    pcn
0  2.454514  2.628551  PID_1
1  2.680562  2.946618  PID_1
2  3.058641  3.524739  PID_1
3  5.835223  6.093277  PID_1
4  6.207301  6.371335  PID_1

vspan = voicing[(voicing["pcn"] == p)]
voiced_mask = mask_from_spans(vspan, env["time_s"].to_numpy())
print(f"{voiced_mask.mean():.0%} voiced")

99% voiced

voicing_segments <- function(f0_path) {
  df <- read_csv(f0_path, show_col_types = FALSE)
  t <- df$time_ms / 1000
  voiced <- !is.na(df$f0)
  edges <- diff(as.integer(voiced))
  onsets  <- t[-1][edges ==  1]
  offsets <- t[-1][edges == -1]
  if (voiced[1]) onsets <- c(t[1], onsets)
  if (voiced[length(voiced)]) offsets <- c(offsets, t[length(t)])
  tibble(onset = onsets, offset = offsets)
}

f0_files <- list.files(acoustics, pattern = "^f0_.*\\.csv$", full.names = TRUE)
head(voicing_segments(f0_files[1]))

# A tibble: 6 × 2
  onset offset
  <dbl>  <dbl>
1  2.45   2.63
2  2.68   2.95
3  3.06   3.52
4  5.84   6.09
5  6.21   6.37
6  6.50   6.64

OpenSmile

opensmilefiles = glob.glob(os.path.join(data_path, "*", "Features", "OpenSmile", "*.csv"))

os_sample = opensmilefiles[0]

os_df = pd.read_csv(os_sample)
os_df.head()

                         end  ... seconds_start
0  0 days 00:00:00.033333333  ...      0.000000
1  0 days 00:00:00.066666667  ...      0.033333
2     0 days 00:00:00.100000  ...      0.066667
3  0 days 00:00:00.133333333  ...      0.100000
4  0 days 00:00:00.166666667  ...      0.133333

[5 rows x 71 columns]

Movement

OpenFace

openfacefiles = glob.glob(os.path.join(data_path, "*", "Features", "OpenFace", "*.csv"))

of_sample = openfacefiles[0]

of_df = pd.read_csv(of_sample)
of_df.head()

   frame   face_id   timestamp   confidence  ...   AU25_c   AU26_c   AU28_c   AU45_c
0      1         0       0.000         0.98  ...      0.0      1.0      0.0      0.0
1      2         0       0.033         0.98  ...      0.0      0.0      0.0      0.0
2      3         0       0.067         0.98  ...      0.0      0.0      0.0      0.0
3      4         0       0.100         0.98  ...      0.0      0.0      0.0      0.0
4      5         0       0.133         0.98  ...      0.0      0.0      0.0      0.0

[5 rows x 714 columns]