pvx vs Rubber Band (Technical Comparison)

This document is intentionally non-marketing. Both tools are strong; each has tradeoffs.

1. Scope and Design Philosophy

Topic	pvx	Rubber Band
Primary form	Python toolkit + multiple CLIs + inspectable DSP code	Highly optimized dedicated stretcher/pitch-shifter library/CLI
Priority order	Audio quality first, speed second	High-quality defaults with strong optimization maturity
Extensibility	Easy to modify in Python (src/pvx/core/*)	Strong core engine, but deeper custom algorithm edits are lower-level
Workflow focus	Processing pipelines, CSV control maps, integrated mastering and docs	Best-in-class focused stretch/pitch engine for many production workflows

2. Areas Where pvx Now Adds Distinct Value

1. Hybrid transient engine (--transient-mode hybrid|wsola), mixing PV steady-state with WSOLA transient handling.
2. Stereo/multichannel coherence controls (--stereo-mode + --coherence-strength).
3. Built-in cycle-consistency benchmark runner (benchmarks/run_bench.py) with regression gate support.
4. Rich CLI ergonomics for new users (--preset, --example, --guided).

3. Areas Where Rubber Band Remains Strong

1. Mature optimized implementation tuned over many years.
2. Widely integrated in DAWs and production tools.
3. Excellent default quality without much parameter tuning.

4. Practical Guidance

Use pvx when:

• you need transparent, inspectable DSP implementation details
• you want to compose custom pipelines and control maps quickly
• you need deterministic CPU regression testing in-repo
• you want to bias decisions toward quality preservation before speed

Use Rubber Band when:

• you need a very mature drop-in engine with minimal tuning
• integration context already standardizes on Rubber Band

5. Reproducible Comparison in This Repo

python3 benchmarks/run_bench.py --quick --out-dir benchmarks/out

If Rubber Band is installed, it is included automatically; if not, it is skipped with a clear note in the report.

6. Important Caveat About "Better"

"Better" is material- and task-dependent:

• vocals vs drums
• mild vs extreme stretch
• mono vs wide stereo
• objective metric score vs subjective musical preference

Use the benchmark report as a directional signal, then always audition critical renders.

Attribution