Sonify an f0 contour

Render a fundamental-frequency contour as an audible waveform: an oscillator whose instantaneous pitch follows f0_hz. Useful for listening to a mean tone contour (e.g. a cluster prototype) rather than only seeing it.

Usage

sonify_f0(
  f0_hz,
  fs = 16000,
  dur = 0.7,
  source = c("tone", "complex", "vowel"),
  vowel = c("a", "i", "u"),
  n_harmonics = 12L,
  rolloff = 0.7,
  fade = 0.02,
  amp = 0.9,
  intensity = NULL,
  dyn_range = 30
)

Arguments

f0_hz

Numeric vector of f0 values in Hz (the contour to play). Leading and trailing NAs are trimmed; internal NAs are interpolated.

fs

Output sample rate (Hz). Default 16000.

dur

Playback duration in seconds. Default 0.7.

source

"tone" for a pure sine, "complex" for a sine plus harmonics with amplitude roll-off (warmer), or "vowel" for a source-filter synthesised vowel (glottal source through formant resonators). The pitch is identical across all three; only the timbre differs.

vowel

Vowel quality for source = "vowel": "a", "i" or "u" (each is a formant set).

n_harmonics

Number of harmonics for source = "complex".

rolloff

Per-harmonic amplitude factor for "complex" (0-1).

fade

Raised-cosine fade in/out duration (s), to avoid clicks.

amp

Peak amplitude (0-1).

intensity

Optional per-frame intensity (dB), the same length as f0_hz. When supplied, the rendered waveform's loudness follows it (dB -> linear amplitude), so the tone swells and fades like the original syllable. NULL (default) gives a flat amplitude.

dyn_range

Loudness range (dB) when intensity is used: frames more than this far below the peak are floored, so quiet frames attenuate without becoming inaudible. Default 30.

Value

A 16-bit mono Wave object.

Details

The contour has no inherent duration, so it is time-scaled to dur seconds. Pitch is rendered by phase accumulation, so the glide is smooth.

All three timbres share one pitch generator and differ only in spectral shape:

1. Contour conditioning. Leading/trailing NAs are trimmed and internal gaps linearly interpolated; the contour is then resampled to round(fs * dur) samples and clamped to 40-2000 Hz.

2. Pitch by phase accumulation. The running phase is phase = cumsum(2 * pi * f0 / fs), so sin(phase) has an instantaneous frequency equal to the contour at every sample; the pitch glides continuously with no discontinuities.

3. Source waveform. "tone" is the bare sine sin(phase). "complex" sums harmonics with a geometric roll-off, sum_h rolloff^(h-1) * sin(h * phase) (n_harmonics terms, capped so the highest stays below the Nyquist frequency). "vowel" uses source-filter synthesis (below).

4. Loudness, fade, normalisation. An optional intensity envelope scales amplitude by 10^((dB - max(dB)) / 20) (floored at -dyn_range dB); a raised-cosine fade removes onset/offset clicks; the waveform is peak-normalised to amp and quantised to 16-bit.

Vowel (source-filter) synthesis. Following the linear source-filter theory of speech production (Fant, 1960), a glottal source is passed through a vocal-tract filter. The source is a band-limited harmonic buzz sum_h (1/h) * sin(h * phase) (the 1/h roll-off approximates the spectral tilt of glottal flow). The filter is a cascade of three two-pole resonators, one per formant, each the recursive difference equation y[n] = x[n] + 2*r*cos(theta)*y[n-1] - r^2*y[n-2], with resonance theta = 2*pi*F/fs and bandwidth set by r = exp(-pi*BW/fs) (poles at radius r, angle theta). The three formant centres (Hz) define the vowel:

	F1	F2	F3
"a"	700	1220	2600
"i"	300	2300	3000
"u"	350	800	2400

The formants are static (no transitions) and the result is a stylised vowel timbre, not a reconstruction of the recording's actual vowel.

References

Fant, G. (1960). Acoustic Theory of Speech Production. Mouton.

Klatt, D. H. (1980). Software for a cascade/parallel formant synthesiser. Journal of the Acoustical Society of America, 67(3), 971-995.

Examples

if (FALSE) { # \dontrun{
w <- sonify_f0(c(120, 140, 160, 150, 130), dur = 0.6, source = "complex")
tuneR::play(w)
} # }