Bolt: Optimize mel filterbank application with sparse matrix multiplication

[ysdede]

What: Optimized the mel filterbank application step by implementing sparse matrix multiplication.
Why: The mel filterbank matrix is extremely sparse (approx. 98% zeros), meaning most multiplications in the inner loop were wasted on zeros.
Impact: Reduces computeRawMel time by ~3-5x.
Measurement: Verified with bun test benchmarks (5s audio: 73ms -> 17ms) and isolated benchmarks. No functional changes; output is identical.

---

PR created automatically by Jules for task 17265254322759491908 started by @ysdede

Summary by CodeRabbit

• Performance

• Mel spectrogram computation is now 3–5x faster, improving overall audio processing speed.

• Documentation

• Added performance optimization notes documenting computational efficiency improvements.

[coderabbitai]

📝 Walkthrough

Walkthrough

The changes implement a performance optimization for mel spectrogram computation by precomputing non-zero element bounds in the sparse mel filterbank matrix. This allows the inner loop in computeRawMel to iterate only over relevant frequency bins instead of the entire range, achieving approximately 3–5x speedup. The optimization is documented and tracked with internal fields; no public APIs are affected.

Changes

Cohort / File(s)	Summary
Documentation .jules/bolt.md	Added dated performance optimization note documenting the sparse mel filterbank optimization and achievable 3–5x speedup.
Sparse Mel Filterbank Optimization src/mel.js	Introduced _fbStart and _fbEnd (Int32Array) fields to precompute non-zero bounds per mel bin. Updated computeRawMel to restrict inner loop iteration using precomputed bounds, replacing full frequency bin sweep with targeted range iteration.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

🐰 Sparse matrices dense with zeros we tame,
Precompute bounds—optimization's sweet name,
Loop only where values dare to dwell,
Speedup achieved with a skip and a spell! ✨

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: optimizing mel filterbank application through sparse matrix multiplication to achieve 3-5x speedup.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ 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-sparse-mel-filterbank-optimization-17265254322759491908

---

No actionable comments were generated in the recent review. 🎉

---

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.}