lcsb

Longest Common Sound Block

Find longest common sound block between multiple wav files.

It can be used to find the duration and start point of opening in a drama series soundtracks.

Finding OP in 9 episodes of an anime series takes 5 seconds on my M1 MBP. Each episode is around 22 minutes long in 16bit mono 48000 Hz wav format.

CLI Usage

❯ lcsb --help
Usage: lcsb [options] <input...>

Arguments:
  input                input files

Options:
  -p, --pool <pool>    max pooling pool size (default: 4800)
  -l, --lower <lower>  partition lower bound (in second) (default: 1)
  -u, --upper <upper>  partition upper bound (in second) (default: 180)
  -m, --mode <mode>    'pair' or 'most' (default: "pair")
  -h, --help           display help for command

Library Usage

lcsb

import fs from "node:fs";
import { partition, lcsb } from "lcsb";

const partition_a = partition(fs.readFileSync("a.wav"));
const partition_b = partition(fs.readFileSync("b.wav"));

const lcsb_ab = lcsb(partition_a, partition_b);

console.log(lcsb_ab);

mclsb

import fs from "node:fs";
import { partition, mclsb } from "lcsb";

const partition_a = partition(fs.readFileSync("a.wav"));
const partition_b = partition(fs.readFileSync("b.wav"));
const partition_c = partition(fs.readFileSync("c.wav"));

const most_common = mclsb([partition_a, partition_b, partition_c]);

console.log(most_common);

Time Complexity

Partition

O(N)

• N is the number of samples in a wav file.

LCSB

O(nm)

• n is the number of partitions in the first wav file.
• m is the number of partitions in the second wav file.

n and m are usually small, so it's almost done in instant.