Skip to content

Bolt: Optimize incremental mel spectrogram computation #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ysdede wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Bolt: Optimize incremental mel spectrogram computation #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions .jules/bolt.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
## 2025-02-12 - [Fake Incremental Computation]
**Learning:** The `IncrementalMelSpectrogram` claimed to reuse cached frames but was actually recomputing the entire audio window (O(N)) and then overwriting the cached part.
**Action:** When working on "incremental" or "caching" features, always benchmark the processing time vs data size. True incremental processing should scale with `new_data`, not `total_data`.
6 changes: 3 additions & 3 deletions src/mel.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,7 +290,7 @@ export class MelSpectrogram {
* @param {Float32Array} audio - Mono PCM audio
* @returns {{rawMel: Float32Array, nFrames: number, featuresLen: number}}
*/
computeRawMel(audio) {
computeRawMel(audio, startFrame = 0) {
const N = audio.length;
if (N === 0) return { rawMel: new Float32Array(0), nFrames: 0, featuresLen: 0 };

Expand All @@ -317,7 +317,7 @@ export class MelSpectrogram {
const { _fftRe: fftRe, _fftIm: fftIm, _powerBuf: powerBuf } = this;
const { hannWindow: window, melFilterbank: fb, nMels, twiddles: tw, nFft, nFreqBins, hopLength, logZeroGuard } = this;

for (let t = 0; t < nFrames; t++) {
for (let t = startFrame; t < nFrames; t++) {
const offset = t * hopLength;
for (let k = 0; k < nFft; k++) { fftRe[k] = padded[offset + k] * window[k]; fftIm[k] = 0; }
fft(fftRe, fftIm, nFft, tw);
Expand Down Expand Up @@ -452,7 +452,7 @@ export class IncrementalMelSpectrogram {
const prefixFrames = Math.floor(prefixSamples / this.processor.hopLength);
const safeFrames = Math.max(0, Math.min(prefixFrames - this.boundaryFrames, this._cachedFeaturesLen));

const { rawMel, nFrames, featuresLen } = this.processor.computeRawMel(audio);
const { rawMel, nFrames, featuresLen } = this.processor.computeRawMel(audio, safeFrames);

if (safeFrames > 0 && this._cachedRawMel) {
for (let m = 0; m < this.nMels; m++) {
Expand Down