🎧 Sound pattern recognition library.
Features:
- Match a sample from a local file.
- Match a sample from a sound stream (Now only mp3 radio stream).
Strategies:
- Use a fingerprint hashing mechanism based on Fast Fourier Transform and Accustic Fingerprint Hash.
- Focus on parallel stream chunk fingerprint hashing method.
- Try to offer a simple way to match a given sample against songs in database collection.
- Provide a built-in solution to handle fingerprint storage.
This library offers Redis database support for storing
hashes and finding matches. By using
trait Repository, it is fairly easy to implement this feature in any database. Using a database engine that supports fast hash lookup is highly recommended. The best performance can be achieved on a fairly simple schema within database RAM.
Improvements:
- The weakest point so far is that only one thread can read and write to the database. The main problem is that the hash lookup wasn't showing any significant performance improvement when using parallel threads. Additionally, when the database size increases, parallel calls make hash lookups take longer. This factor made me stick to single-threaded database calls. (If you know how to improve this, please provide your solution or issue a proposal ticket).
- Test against self cached file samples
- Run local redis database with
systemctl start redis-server.service - Test it with
cargo test -- --nocaptureorcargo test<- with less information
- Building. Please build with
RUSTFLAGS="--emit=asm"flag enabled, which forces the compiler to use a single LLVM module for the entire crate, allowing better LLVM optimization.
Hardware: Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz, Intel Corporation Cannon Point-LP Shared SRAM Speed: 2400 MT/s, Build with: rust stable-x86_64-unknown-linux-gnu. Redis server: v=5.0.7 sha=00000000:0 malloc=jemalloc-5.2.1 bits=64 build=49eb440ab0786b6a
STATISTICS:
Total time to calculate and populate db with 31 songs took: 142626 ms Average time to calculate and add to db 1 song 4600 ms Average time to calculate and add to db 1000 fingerprints took: 807 ms Calculating samples fingerprints
Track : Sample Guns N' Roses - Welcome To The Jungle.mp3 : Guns N' Roses - Welcome To The Jungle.mp3 Match accuracy 58 %
Track : Sample Maroon 5 - Memories.mp3 : Maroon 5 - Memories.mp3 Match accuracy 61 %
Track : Sample Tove Lo - Habits .mp3 : Tove Lo - Habits .mp3 Match accuracy 70 %
Track : Sample twenty one pilots - Ride.mp3 : twenty one pilots - Ride.mp3 Match accuracy 62 %
Total time to calculate and match 4 samples took: 2430 ms Average time to calculate fingerprint for 1 sample and match it against fingerprints in db took 607 ms
All samples matched corresponding songs.
🌸 Dla Grzechotki 🍒