Bolt: Optimize mel filterbank matrix multiplication

[ysdede]

What: Implemented sparse iteration for the Mel filterbank matrix multiplication.
Why: The Mel filterbank is ~98% sparse (mostly zeros). Iterating over the full frequency range for each mel bin was performing unnecessary multiplications by zero.
Impact: Reduces processing time by ~3x (134ms -> 46ms for 10s audio).
Measurement: Verified with a custom benchmark script and existing ONNX correctness tests.

---

PR created automatically by Jules for task 5107709169723486900 started by @ysdede

Summary by CodeRabbit

• Documentation

• Added documentation describing Mel filterbank optimization strategy leveraging sparsity patterns.

• Refactor

• Optimized Mel filterbank computation by precomputing non-zero index ranges, achieving approximately 3x performance improvement.

[coderabbitai]

📝 Walkthrough

Walkthrough

Documentation and implementation of a Mel filterbank optimization strategy. The changes precompute start and end indices for non-zero values in sparse filterbanks, enabling the filterbank computation to skip zero multiplications during mel-spectrogram processing.

Changes

Cohort / File(s)	Summary
Documentation .jules/bolt.md	Adds optimization proposal describing ~98% sparsity in Mel filterbanks and suggests precomputing non-zero index ranges for ~3x speedup.
Implementation src/mel.js	Implements precomputation of _fbStart and _fbEnd arrays during initialization; refactors loops in computeRawMel and main processing to use index ranges instead of full filterbank iterations.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐰 A sparsity so grand, ninety-eight zeros stand,
We skip them with grace, precomputed in place,
Three times faster now, filterbanks take a bow,
Index ranges so neat, make optimization complete! 🚀

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly addresses the main optimization—sparse iteration for mel filterbank matrix multiplication—which is the primary change in this PR.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Merge Conflict Detection	✅ Passed	✅ No merge conflicts detected when merging into main

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch bolt-mel-sparsity-optimization-5107709169723486900

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In @.jules/bolt.md:
- Around line 1-3: Update the documentation header date in .jules/bolt.md (the
line starting with "## 2024-05-23 - Mel Filterbank Sparsity") to the correct PR
date (February 2026) or the actual change date; edit that header so it reads the
accurate date (e.g., "## 2026-02-XX - Mel Filterbank Sparsity") to keep the
changelog consistent with the PR.

🧹 Nitpick comments (1)

src/mel.js (1)

264-291: Edge case: an all-zero mel bin row will silently fall back to a full-range scan.

If every filterbank value for a given mel bin is zero (unlikely but possible with unusual nMels/sampleRate/nFft combos), neither break fires, so start stays 0 and end stays this.nFreqBins — the loop degrades to the old full-range behavior. This is functionally safe but worth a note.

You could also tighten it by initializing end = 0 so that the inner loop becomes a no-op for empty rows:

Suggested tweak

     for (let m = 0; m < this.nMels; m++) {
-      let start = 0;
-      let end = this.nFreqBins;
+      let start = this.nFreqBins;  // default: empty range
+      let end = 0;
       const fbOff = m * this.nFreqBins;
 
       // Find first non-zero

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Stale date in the documentation header.

The date 2024-05-23 doesn't match the PR date (February 2026). Consider updating it to reflect the actual date of the change.

🤖 Prompt for AI Agents

In @.jules/bolt.md around lines 1 - 3, Update the documentation header date in
.jules/bolt.md (the line starting with "## 2024-05-23 - Mel Filterbank
Sparsity") to the correct PR date (February 2026) or the actual change date;
edit that header so it reads the accurate date (e.g., "## 2026-02-XX - Mel
Filterbank Sparsity") to keep the changelog consistent with the PR.