Introduction to whatsappaudioR

Reading a voice note

library(whatsappaudioR)

w <- wa_read("voice_note.opus")
w

wa_read() detects the file extension and, for compressed formats, decodes via a system ffmpeg binary into a tuneR::Wave object – the same representation you’d get from tuneR::readWave() on a plain .wav file, so the rest of the R audio ecosystem (tuneR, seewave) works on it unchanged.

Tidy spectrogram

spec_df <- wa_spectrogram(w, wl = 1024)
head(spec_df)

#> # A tibble: 6 x 3
#>    time  freq amplitude
#>   <dbl> <dbl>     <dbl>
#> 1 0.032 0.000     -38.2
#> 2 0.032 0.016     -41.7
#> ...

freq is in kHz (matching seewave’s convention); amplitude is in dB. Because the result is a long-format tibble, it plots directly with ggplot2:

library(ggplot2)

ggplot(spec_df, aes(x = time, y = freq, fill = amplitude)) +
  geom_raster() +
  scale_fill_viridis_c() +
  labs(x = "Time (s)", y = "Frequency (kHz)", fill = "Amplitude (dB)")

Spectral features

wa_features(w)

Returns a one-row tibble of documented spectral features (peak frequency, spectral centroid, flatness, entropy, rolloff, dominant-frequency stability, a harmonic-energy heuristic, and 12 MFCCs) – see ?wa_features for exact definitions. Nothing here is a black box: every column is a named, inspectable, conventional bioacoustic/audio feature.

Simulating compression

If you have clean, uncompressed audio and want to know how it would sound (and measure) after passing through a WhatsApp-style pipeline:

clean <- wa_read("clean_recording.wav")
degraded <- wa_simulate_compression(clean, bitrate = 16000, sr = 16000)

This requires a system ffmpeg build with libopus support.

Batch processing

features_df <- wa_batch("voice_notes/", wa_features)

Applies any function of a Wave object (default wa_features()) across every supported audio file in a directory, returning one row per file with a file column.