needle

A tool that finds a needle (opening/intro and ending/credits) in a haystack (TV or anime episode).

Demo

Quickstart

Download and run the needle Docker container:

docker run ghcr.io/aksiksi/needle:latest --help

Note: If you just want to install the binary, skip over to the install section before proceeding.

Run a search for opening and endings in the first three episodes of Land of the Lustrous:

$ needle search --analyze ~/Movies/land-of-lustrous-ep1.mkv ~/Movies/land-of-lustrous-ep2.mkv ~/Movies/land-of-lustrous-ep3.mkv

~/Movies/land-of-lustrous-ep1.mkv

* Opening - N/A
* Ending - "22:10s"-"23:39s"

~/Movies/land-of-lustrous-ep2.mkv

* Opening - "00:43s"-"02:12s"
* Ending - "22:10s"-"23:56s"

~/Movies/land-of-lustrous-ep3.mkv

* Opening - "00:40s"-"02:08s"
* Ending - "22:09s"-"23:56s"

Run the same search as above, but write the results to a JSON file stored alongside each video (called a "skip file"):

$ needle search --analyze --no-display --write-skip-files ~/Movies/land-of-lustrous-ep1.mkv ~/Movies/land-of-lustrous-ep2.mkv ~/Movies/land-of-lustrous-ep3.mkv

$ cat ~/Movies/land-of-lustrous-ep1.needle.skip.json
{"opening":null,"ending":[1331.6644,1419.0249],"md5":"14bfa97f85d86f74e1ab5a26066f9181"}%

Overview

needle has two subcommands: 1) analyze and 2) search.

You may have noticed that we only used the search subcommand in the examples above. You also likely noticed that it takes quite a bit to of time to spit out results. Well, it turns out that decoding and resampling audio streams takes way longer than searching for openings and endings.

That's where the analyze command comes in. Using this subcommand, you can pre-compute the required data and store it alongside video files (just like with skip files). The pre-computed data is stored in a compact binary format and is much smaller in size than the audio stream.

Let's try it out with the same files as above:

$ needle analyze ~/Movies/land-of-lustrous-ep1.mkv ~/Movies/land-of-lustrous-ep2.mkv ~/Movies/land-of-lustrous-ep3.mkv

$ ls -la ~/Movies/land-of-lustrous-*.needle.dat
-rw-r--r--  1 aksiksi  staff  76128 Jul  2 20:09 ~/Movies/land-of-lustrous-ep1.needle.dat
-rw-r--r--  1 aksiksi  staff  76128 Jul  2 20:09 ~/Movies/land-of-lustrous-ep2.needle.dat
-rw-r--r--  1 aksiksi  staff  76128 Jul  2 20:09 ~/Movies/land-of-lustrous-ep3.needle.dat

The frame hash files are quite small: on the order of 4 KB per minute of audio. Note that the size will change based on how you configure the analyzer.

Once we have these pre-computed files, we can re-run the search step, but this time we can omit the --analyze flag:

$ needle search ~/Movies/land-of-lustrous-ep1.mkv ~/Movies/land-of-lustrous-ep2.mkv ~/Movies/land-of-lustrous-ep3.mkv

~/Movies/land-of-lustrous-ep1.mkv

* Opening - N/A
* Ending - "22:10s"-"23:39s"

~/Movies/land-of-lustrous-ep2.mkv

* Opening - "00:43s"-"02:12s"
* Ending - "22:10s"-"23:56s"

~/Movies/land-of-lustrous-ep3.mkv

* Opening - "00:40s"-"02:08s"
* Ending - "22:09s"-"23:56s"

On my machine (M1 Macbook Pro), the analyze step takes 10 seconds, while the search using pre-computed data takes less than 1 second.

Let's try running analyze and search for Season 4 of Attack on Titan (yes, you can specify directories!):

$ time needle analyze ~/Movies/Season\ 04
needle analyze ~/Movies/Season\ 04 --force  87.37s user 6.51s system 783% cpu 11.983 total

$ time needle search ~/Movies/Season\ 04
needle search ~/Movies/Season\ 04  32.59s user 8.82s system 737% cpu 5.617 total

In this case, the search step ran about 2x faster than the analyze step. Note that needle utilizes all available cores by default (via rayon), but you can disable this by passing in --no-threading. For reference, this directory contains 28 episodes of Attack on Titan (~20 minutes each).

Configuration

TODO

Install

There are currently three ways to install needle:

Run needle in a Docker container (amd64 only!):

docker run ghcr.io/aksiksi/needle:latest --help

Grab the latest binary from the releases page
- Prerequisites: Install the runtime dependencies for your platform
Build and install the latest release:
- Prerequisites: Install the build dependencies for your platform

cargo install needle-rs

Build latest version (HEAD) from source:
- Prerequisites: Install the build dependencies for your platform

cargo build --release --manifest-path needle/

Runtime Dependencies

Linux (Debian/Ubuntu)

Download the FFmpeg libraries:

sudo apt-get install \
    libfftw3-3 \
    libavutil56 \
    libavformat58 \
    libswresample3 \
    libavcodec58

macOS

Install FFmpeg and libraries:

brew install ffmpeg

Build

Linux (Debian/Ubuntu)

Install pkg-config, cmake, libclang, fftw3 (optional, but recommended), and the FFmpeg libraries:

sudo apt-get install \
    pkg-config \
    cmake \
    libclang-dev \
    libfftw3-dev \
    libavutil-dev \
    libavformat-dev \
    libswresample-dev \
    libavcodec-dev

Build:

cargo install --path .

This will dynamically link against FFmpeg and statically link chromaprint.

Dynamic

Install libraries:

sudo apt-get install \
    pkg-config \
    cmake \
    libclang-dev \
    libfftw3-dev \
    libavutil-dev \
    libavformat-dev \
    libswresample-dev \
    libavcodec-dev

Build:

CHROMAPRINT_SYS_DYNAMIC=1 cargo install --path .

macOS

Install cmake and FFmpeg:

brew install cmake pkg-config ffmpeg

Build:

cargo install --path .

This will dynamically link against FFmpeg. chromaprint will be statically linked.

Windows

Install cargo-vcpkg:

cargo install cargo-vcpkg

Install vcpkg deps:

cargo vcpkg build

Build:

# Statically link against both FFmpeg and chromaprint
cargo build --release --features static

Dynamic

Set the following environment variables:

a. To dynamically link both FFmpeg and chromaprint:

# Powershell
$env:VCPKGRS_DYNAMIC='1'
$env:VCPKGRS_TRIPLET='x64-windows'

# Git bash
export VCPKGRS_DYNAMIC=1
export VCPKGRS_TRIPLET='x64-windows'

b. Just chromaprint:

# Powershell
$env:CHROMAPRINT_SYS_DYNAMIC='1'

# Git bash
export CHROMAPRINT_SYS_DYNAMIC=1

Build deps:

cargo vcpkg build

Build needle:

cargo build --release

License

This work is dual-licensed under MIT and LGPL 2.1 (or later).

If you choose to statically link FFmpeg, this is licensed as LGPL 2.1 (or later) due to FFmpeg. Otherwise, you can use the MIT license.

SPDX-License-Identifier: MIT OR LGPL-2.1-or-later