fast-phase-vocoder

A pure JavaScript, zero-dependency, transient-preserving phase vocoder for high-quality audio time stretching.

Why?

Most web developers looking for high-quality audio time-stretching (changing playback speed without altering pitch) are forced to use heavy WebAssembly (WASM) ports like rubberband-wasm. WASM has boot-time overhead and complicates build pipelines.

This library is a 100% pure JavaScript implementation of a phase vocoder that locks phases on transients (sharp sounds like consonants). This prevents the "metallic" or "echoey" artifacts common in naive JS stretchers.

• Fast: Processes 10 seconds of 24kHz audio in <100ms (100x faster than real-time).
• Zero Dependencies: No WASM, no external NPM packages required.
• Browser & Node: Works perfectly in AudioContext or raw Float32Arrays.

Installation

npm install fast-phase-vocoder

Audio Samples

Listen to the phase vocoder in action. These samples were generated natively using fast-phase-vocoder:

• Baseline (1.0x Speed)
• Slowed Down (0.8x Speed)
• Sped Up (1.5x Speed)
• Fast (2.0x Speed)

Usage

In the Browser (Web Audio API)

import { retimeAudioBuffer } from 'fast-phase-vocoder';

// Assuming you have an AudioContext and a decoded AudioBuffer
const newSpeed = 1.5; // 1.5x speed
const retimedBuffer = retimeAudioBuffer(audioContext, originalAudioBuffer, newSpeed);

const source = audioContext.createBufferSource();
source.buffer = retimedBuffer;
source.connect(audioContext.destination);
source.start();

Raw Float32Array / PCM

import { retimeMonoPcm } from 'fast-phase-vocoder';

const pcm = new Float32Array([...]); // Your raw mono audio data
const sampleRate = 24000;
const newSpeed = 0.8; // Slow down to 80%

const retimedPcm = retimeMonoPcm(pcm, sampleRate, newSpeed);

Credits

Mathematical processing (FFT) adapted from HuggingFace Transformers JS.