umt_resampler

Lightweight, high-performance, header-only audio resampler with fused channel mixing, designed for real-time and streaming applications.

---

Overview

umt_resampler is a minimal, dependency-free C++ audio resampler built for low-latency systems.
It implements a polyphase FIR resampler with Kaiser windowing and integrates channel mixing in the same processing stage for maximum efficiency.

The implementation is written in clean scalar C++ to allow compilers (Clang/GCC) to auto-vectorize (SIMD) without requiring platform-specific intrinsics.

---

Features

• Real-time and streaming friendly

  • No dynamic allocations in the processing path
  • Deterministic performance

• High-performance polyphase FIR resampling

  • Efficient sample rate conversion (e.g. 48kHz → 44.1kHz)
  • Kaiser window filter design

• Configurable quality

  • Supports 16-tap and 32-tap modes
  • Default: 16 taps (balanced performance/quality)

• Fused channel mixing

  • Multi-channel input/output support
  • Layout-aware processing (e.g. stereo, surround)
  • Efficient downmix (e.g. 5.1 / 7.1 → stereo)

• Float processing

  • Input: float
  • Output: float
  • Expected range: [-1.0, 1.0]

• Auto-vectorization friendly

  • Scalar implementation optimized for compiler SIMD generation (NEON/SSE)

• Header-only

  • No external dependencies
  • Requires only standard C/C++ library

---

Basic Example

#include "umt_resampler.hpp"

umt::Resampler rs;
rs.init(2, 2, 48000, 44100);

float input[960];   // 480 frames stereo
float output[960];

uint32_t consumed = 0;

uint32_t produced = rs.resample(
    input,
    960,
    output,
    consumed
);

Streaming Example (Recommended)

#include "umt_resampler.hpp"
#include <vector>
#include <cstdint>

int main()
{
    const int srcChannels = 2;
    const int dstChannels = 2;
    const int srcRate = 48000;
    const int dstRate = 44100;

    umt::Resampler rs;
    rs.init(srcChannels, dstChannels, srcRate, dstRate);

    const int chunkFrames = 256;
    const int chunkSamples = chunkFrames * srcChannels;

    std::vector<float> input(chunkSamples);

    // Fill with dummy data (replace with real audio)
    for (auto &v : input) v = 0.1f;

    // Allocate correct output size
    uint32_t maxOut = rs.calculateOutputBufferSize(chunkSamples);
    std::vector<float> output(maxOut);

    uint32_t consumedTotal = 0;

    while (consumedTotal < chunkSamples)
    {
        uint32_t consumed = 0;

        uint32_t produced = rs.resample(
            input.data() + consumedTotal,
            chunkSamples - consumedTotal,
            output.data(),
            consumed
        );

        consumedTotal += consumed;

        if (consumed == 0)
            break; // safety guard

        // process 'produced' samples here
        // e.g. send to audio device or encoder
    }

    return 0;
}

Benchmark

Config	Value
Channels	2 → 2
Sample Rate	48000 → 44100
Samples/sec	112250239.13
Frames/sec	56125119.57
Realtime Factor	1272.68x
Last Updated	2026-04-29 08:59 UTC