Skip to content

mayadem8/mp3_match

Repository files navigation

MP3-in-MP3 (GPU)

Installation

Clone the repository

https://github.com/mayadem8/mp3_match.git
cd mp3_match

Download ffmpeg: https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-essentials.7z

Put ffmpeg.exe in mp3_match folder

Download CUDA Toolkit Installer: https://developer.download.nvidia.com/compute/cuda/13.0.2/local_installers/cuda_13.0.2_windows.exe

Update Nvidia driver, so it supports CUDA 13, can be found here: https://www.nvidia.com/en-us/drivers/

Install dependencies

pip install -r requirements.txt

Put all samples in samples folder and all chunks in broadcasts folder:

image

Aftre you need to create .npy databases for all audios, run:

python .\build_samples_base.py   
python .\build_broadcasts_base.py   

Run main code

python .\mp3match_gpu.py

Example Output:

================  BROADCAST: chunk_20250127_135809.mp3  ================

GPU ready for broadcast (len=60.0 min) in 0.39s

=== MATCH SUMMARY ===

✅ GPU BATCH COMPLETE in 0.55s

გაგრძელება იქნება.mp3 →   164.10s  ( 99.93%)
გრძონების აღზრდა.mp3 →    56.80s  ( 99.96%)
ზიმფერი.mp3          →    37.00s  ( 99.96%)
მოამბე.mp3           →  3369.50s  ( 99.93%)
მოწევა მავნებელია1.mp3 →   177.80s  ( 99.98%)
მოწევა მავნებელია2.mp3 →  3332.70s  ( 99.97%)
ნინა სეტენტრე.mp3    →   125.80s  ( 99.94%)
პრაქტიკანტი.mp3      →   186.50s  ( 99.96%)
პრეზიდენტი ლაბლე.mp3 →     7.30s  ( 99.99%)
სანდომი.mp3          →    27.20s  ( 99.92%)
ტრი ჟელანია.mp3      →    47.00s  ( 99.94%)
ფინო.mp3             →    17.40s  ( 99.94%)
წიგნების თარო.mp3    →    92.00s  ( 99.97%)

================  BROADCAST: chunk_20250127_145802.mp3  ================

GPU ready for broadcast (len=60.0 min) in 0.15s

=== MATCH SUMMARY ===

✅ GPU BATCH COMPLETE in 0.89s

გაგრძელება იქნება.mp3 →  2968.11s  (  5.15%)
გრძონების აღზრდა.mp3 →  2994.13s  (  2.46%)
ზიმფერი.mp3          →  3418.24s  (  5.83%)
მოამბე.mp3           →  2953.12s  ( 23.38%)
მოწევა მავნებელია1.mp3 →  2185.71s  (  7.99%)
მოწევა მავნებელია2.mp3 →  3040.13s  ( 50.36%)
ნინა სეტენტრე.mp3    →  2972.32s  (  2.64%)
პრაქტიკანტი.mp3      →  1986.20s  (  1.93%)
პრეზიდენტი ლაბლე.mp3 →  3476.42s  (  5.28%)
სანდომი.mp3          →  2966.53s  (  6.94%)
ტრი ჟელანია.mp3      →  3452.49s  (  5.50%)
ფინო.mp3             →  2953.19s  (  5.77%)
წიგნების თარო.mp3    →  2932.64s  (  2.35%)

.
.
.

ALL BROADCASTS COMPLETE in 17.37s

About

A high-performance Python application for detecting duplicate or related MP3 recordings in real time. It uses FFmpeg for audio decoding and CuPy (CUDA) for GPU-based similarity search, enabling massive speed-ups over CPU processing. Optimized and tested on AWS GPU instances (Tesla M40/L40) for scalable, production-grade audio analysis.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages