Spelling, signals and block columns
Audio signals in one jsonW row
From 16-bit PCM, jsonW measures pitch, pauses, intensity, jitter, shimmer, HNR, formants, MFCC, spectral centroid, articulation rate and breaths.
· jsonW 1.3.11
The problem
Measuring how someone spoke means combining phonetics and signal processing tools, each with its own output, before storing the result anywhere.
How jsonW solves it
The jsonw_sinal_audio function takes 16-bit mono PCM audio and returns one jsonW-text row in the declared schema. Pitch comes from normalized autocorrelation, the same method Praat uses; jitter and shimmer from glottal pulses; formants from LPC; and pauses from Otsu's threshold, which takes the cut-off from the data itself.
int64_t r = jsonw_sinal_audio(pcm, n, 16000, "sintetico", "a1", saida, sizeof saida, erro, sizeof erro);
{origem: "sintetico", chave: "a1", dur_s: 2, fala_s: 1.6, silencio_pct: 19.1919, pausas: 1, pausa_media_ms: 380, pausa_max_ms: 380, nucleos: 1, taxa_articulacao: 0.625, taxa_fala: 0.5, intensidade_db: -16.0177, intensidade_dp: 0.74886, f0_hz: 164.993, f0_dp: 14.9999, f0_min: 149.997, f0_max: 179.998, jitter_pct: 0.00237075, shimmer_pct: 0.123395, hnr_db: 43.8417, f1: 624.1, f2: 1124.45, f3: 3245.8, mfcc: vf32[-16.138, 14.372, 2.5701, -5.5633, -3.0267, 1.5794, 2.0231, -0.98157, -2.1028, -0.48222, 0.94709, 0.17767, -1.0651], centroide_hz: 293.164, respiracoes: 0, disfluencias: null, vocalizacoes: null, vozeados: 156, quadros: 160}
Real libjsonw 1.3.11 output for 2 s of synthetic audio: voice at 150 Hz, 0.4 s of silence, voice at 180 Hz. It found 1 pause of 380 ms and the pitch between 150 and 180 Hz.
Where it is
In the ABI (jsonw_sinal_audio, with the schema in jsonw_sinal_esquema). The row can be written straight into a .jsw file. The feature measures, it does not decide: your model interprets the numbers.