@audio/mir-tempogram 1.0.0

package/package.json

@@ -0,0 +1,32 @@
+{
+  "name": "@audio/mir-tempogram",
+  "version": "1.0.0",
+  "description": "Tempogram \u2014 local tempo over time via onset-envelope autocorrelation",
+  "type": "module",
+  "sideEffects": false,
+  "main": "tempogram.js",
+  "exports": {
+    ".": "./tempogram.js",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "tempogram.js"
+  ],
+  "dependencies": {
+    "@audio/beat-core": "^1.0.0"
+  },
+  "keywords": [
+    "audio",
+    "dsp",
+    "mir",
+    "tempogram"
+  ],
+  "license": "MIT",
+  "author": "Dmitry Iv. <dfcreative@gmail.com>",
+  "engines": {
+    "node": ">=18"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}

package/tempogram.js

@@ -0,0 +1,41 @@
+// Tempogram — local tempo over time: onset envelope (spectral flux) autocorrelated
+// over sliding windows. Shorter lags accumulate more terms, biasing toward the
+// notated tempo over its subdivisions (librosa/MIREX tempogram class).
+
+import { spectralFlux } from '@audio/beat-core'
+
+/**
+ * @param {Float32Array} data — mono PCM
+ * @param {object} opts — { fs=44100, frameSize=2048, hopSize=512,
+ *   window=6 (s), hop=2 (s), minBpm=40, maxBpm=240 }
+ * @returns {{ times: Float32Array, bpm: Float32Array, matrix: Float32Array[],
+ *   odfRate: number, minLag: number, maxLag: number }}
+ */
+export default function tempogram (data, { fs = 44100, frameSize = 2048, hopSize = 512, window = 6, hop = 2, minBpm = 40, maxBpm = 240 } = {}) {
+	let { odf } = spectralFlux(data, { fs, frameSize, hopSize })
+	let odfRate = fs / hopSize
+	let winN = Math.round(window * odfRate)
+	let hopN = Math.max(1, Math.round(hop * odfRate))
+	let minLag = Math.max(1, Math.round(odfRate * 60 / maxBpm))
+	let maxLag = Math.round(odfRate * 60 / minBpm)
+
+	let times = [], bpms = [], matrix = []
+	for (let s = 0; s + winN <= odf.length; s += hopN) {
+		let mean = 0
+		for (let i = s; i < s + winN; i++) mean += odf[i]
+		mean /= winN
+
+		let row = new Float32Array(maxLag - minLag + 1)
+		let best = -Infinity, bestLag = minLag
+		for (let lag = minLag; lag <= maxLag; lag++) {
+			let acc = 0
+			for (let i = s + lag; i < s + winN; i++) acc += (odf[i] - mean) * (odf[i - lag] - mean)
+			row[lag - minLag] = acc
+			if (acc > best) { best = acc; bestLag = lag }
+		}
+		matrix.push(row)
+		times.push((s + winN / 2) / odfRate)
+		bpms.push(60 * odfRate / bestLag)
+	}
+	return { times: Float32Array.from(times), bpm: Float32Array.from(bpms), matrix, odfRate, minLag, maxLag }
+}