owen/lms
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Speed up audio similarity engine startup for big collections

Browse Source
This commit is contained in:
emeric
2026-06-07 10:21:57 +02:00
parent 322362b3b9
commit 17ca039087
1 changed files with 16 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/libs/services/recommendation/impl/audio-similarity/AudioSimilarityEngine.impl.hpp
+16 -10
View File
@@ -31,7 +31,6 @@
#include "core/ILogger.hpp" #include "core/ILogger.hpp"
#include "core/ITraceLogger.hpp" #include "core/ITraceLogger.hpp"
#include "core/Random.hpp"
#include "database/IDb.hpp" #include "database/IDb.hpp"
#include "database/Session.hpp" #include "database/Session.hpp"
@@ -682,20 +681,26 @@ namespace lms::recommendation
LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "computeTrackDistanceThreshold"); LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "computeTrackDistanceThreshold");
constexpr std::size_t maxSampleCount{ 500 }; constexpr std::size_t maxSampleCount{ 500 };
constexpr std::size_t maxCandidateCount{ 10'000 };
constexpr float stdDevMultiplier{ 2.F }; constexpr float stdDevMultiplier{ 2.F };
const std::size_t sampleCount{ std::min(_trackVectors.size(), maxSampleCount) };
LOG(INFO, "computing track distance threshold using " << sampleCount << " samples...");
// Collect all vector pointers and shuffle for an unbiased random sample.
std::vector<const ReducedVector*> allVectors; std::vector<const ReducedVector*> allVectors;
allVectors.reserve(_trackVectors.size()); allVectors.reserve(_trackVectors.size());
for (const auto& [id, vec] : _trackVectors) for (const auto& [id, vec] : _trackVectors)
allVectors.push_back(vec); allVectors.push_back(vec);
// move maxCandidateCount random elements to the front
const std::size_t candidateCount{ std::min(allVectors.size(), maxCandidateCount) };
std::minstd_rand randomEngine{ 42 }; std::minstd_rand randomEngine{ 42 };
core::random::shuffleContainer(randomEngine, allVectors); for (std::size_t i{}; i < candidateCount; ++i)
{
std::uniform_int_distribution<std::size_t> dist{ i, allVectors.size() - 1 };
std::swap(allVectors[i], allVectors[dist(randomEngine)]);
}
allVectors.resize(candidateCount);
const std::size_t sampleCount{ std::min(candidateCount, maxSampleCount) };
LOG(INFO, "computing track distance threshold using " << sampleCount << " samples on " << candidateCount << " candidates...");
math::StatsAccumulator<FloatType> stats; math::StatsAccumulator<FloatType> stats;
for (std::size_t i{}; i < sampleCount; ++i) for (std::size_t i{}; i < sampleCount; ++i)
@@ -714,7 +719,8 @@ namespace lms::recommendation
minDist = d; minDist = d;
} }
stats.add(minDist); if (minDist < std::numeric_limits<FloatType>::max())
stats.add(minDist);
} }
if (stats.getCount() >= 2) if (stats.getCount() >= 2)
@@ -731,7 +737,7 @@ namespace lms::recommendation
LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "ComputeReleaseDistanceThreshold"); LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "ComputeReleaseDistanceThreshold");
constexpr std::size_t maxSampleCount{ 200 }; constexpr std::size_t maxSampleCount{ 200 };
constexpr std::size_t maxCandidateCount{ 1'000 }; constexpr std::size_t maxCandidateCount{ 2'000 };
constexpr float stdDevMultiplier{ 2.F }; constexpr float stdDevMultiplier{ 2.F };
using CosineDistance = math::NormalizedCosineDistance<ReducedVector::getSize(), FloatType>; using CosineDistance = math::NormalizedCosineDistance<ReducedVector::getSize(), FloatType>;
@@ -784,7 +790,7 @@ namespace lms::recommendation
LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "ComputeArtistDistanceThreshold"); LMS_SCOPED_TRACE_DETAILED("AudioSimilarityEngine", "ComputeArtistDistanceThreshold");
constexpr std::size_t maxSampleCount{ 200 }; constexpr std::size_t maxSampleCount{ 200 };
constexpr std::size_t maxCandidateCount{ 1'000 }; constexpr std::size_t maxCandidateCount{ 2'000 };
constexpr float stdDevMultiplier{ 2.F }; constexpr float stdDevMultiplier{ 2.F };
using CosineDistance = math::NormalizedCosineDistance<ReducedVector::getSize(), FloatType>; using CosineDistance = math::NormalizedCosineDistance<ReducedVector::getSize(), FloatType>;