Integration between echoprint-codegen and Java

README.md

@@ -59,16 +59,47 @@ The makefile builds an example code generator that uses libcodegen, called "code
 
 Will take 30 seconds of audio from 10 seconds into the file and output JSON suitable for querying:
 
-    {"metadata":{"artist":"Michael jackson", "release":"800 chansons des annes 80", "title":"Billie jean", "genre":"", "bitrate":192, "sample_rate":44100, "seconds":294, "filename":"billie_jean.mp3", "samples_decoded":220598, "given_duration":30, "start_offset":10, "version":4.00}, "code_count":846, "code":"JxVlIuNwzAMQ1fxCDL133+xo1rnGqNAEcWy/ERa2aKeZmW...
+    [
+    {"metadata":{"artist":"Michael Jackson", "release":"Thriller", "title":"Billie Jean", "genre":"", "bitrate":128,"sample_rate":44100, "duration":294, "filename":"billie_jean.mp3", "samples_decoded":330902, "given_duration":30, "start_offset":10, "version":4.12, "codegen_time":0.087329, "decode_time":0.297166}, "code_count":906, "code":"eJztmm2OZacORafEt2E4YGD-Q8jCt1UnXdKlIlVa ..."
+    ]
 
-You can host your own [Echoprint server](http://github.com/echonest/echoprint-server "echoprint-server") and ingest or query to that.
+You can POST this JSON directly to the Echo Nest's [song/identify](http://developer.echonest.com/docs/v4/song.html#identify "song/identify") (who has an Echoprint server booted), for example:
+
+    curl -F "query=@post_string" http://developer.echonest.com/api/v4/song/identify?api_key=YOUR_KEY
+    {"response": {"status": {"version": "4.2", "code": 0, "message": "Success"}, "songs": [{"tag": 0, "score": 273, "title": "Billie Jean", "message": "OK (match type 6)", "artist_id": "ARXPPEY1187FB51DF4", "artist_name": "Michael Jackson", "id": "SOJIZLV12A58A78309"}]}}
+    (you can also use GET, see the API description)
+
+Or you can host your own [Echoprint server](http://github.com/echonest/echoprint-server "echoprint-server") and ingest or query to that.
 
 Codegen also runs in a multithreaded mode for bulk resolving:
 
     ./echoprint-codegen -s 10 30 < file_list
 
 Will compute codes for every file in file_list for 30 seconds starting at 10 seconds. (It tries to be smart about the number of threads to use.) It will output a JSON list.
 
+## Integration with Scala and Java via JNI
+You can use echoprint-codegen inside a JVM. For this you need to create a class named Echoprint in package com.playax.fingerprint. Here is a Scala example of this class:
+
+    package com.playax.fingerprint
+
+    class Echoprint {
+       @native def code(fileName: String): String 
+    }
+
+    object Echoprint {
+      val EP = new Echoprint
+
+      System.load("/path/to/libcodegen.4.1.2.dylib")
+
+     def code(fileName: String) = EP.code(fileName)
+    }
+
+Then you call static Method:
+
+    Echoprint.code("/path/to/file.mp3")
+
+And it will return the json with fingerprint data.
+
 ## Statistics
 
 ### Speed

build-debian-buster.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+# 
+# Simple recipe for building echoprint-codegen on Debian Buster. Called
+# from our Cloudbuild config.
+set -e
+
+apt update
+apt install -y build-essential ffmpeg libboost1.67-dev libtag1-dev zlib1g-dev 
+
+# This thing needs JNI headers to be built, even if you're building only the CLI. 
+# Fortunately for us it compiles with the version in Buster.
+apt install -y openjdk-11-jdk
+export JAVA_HOME="/usr/lib/jvm/java-11-openjdk-amd64/"
+
+cd src
+make

cloudbuild.yml

@@ -0,0 +1,30 @@
+# This config builds echoprint-codegen for a Linux distro and uploads it
+# into GCS. This is NOT a good solution, as the binary will probably break
+# down if libraries in the OS change even by an inch.
+#
+# Ideally, this should be repackaged as a microservice and built as a regular
+# container image deployed as a sidecar, and the main Playax app should access 
+# this as an API instead of expecting to have the binary available locally.
+#
+# Other options would include building an APT or AppImage package, but clearly a 
+# microservice would provide better isolation for dependencies.
+options:
+  dynamic_substitutions: true
+
+substitutions:
+  _DISTRO: 'debian'
+  _RELEASE: 'buster'
+  _BINARIES_BUCKET: 'gs://files.playax.com/binaries'
+  _ARTIFACT_URL: '${_BINARIES_BUCKET}/echoprint-codegen/${_DISTRO}-${_RELEASE}/echoprint-codegen'
+
+steps:
+  - name: '${_DISTRO}:${_RELEASE}'
+    args: ['bash', '-c', './build-${_DISTRO}-${_RELEASE}.sh']
+
+  # We unfortunately cannot use the cloudbuild artifact copy because
+  # we need to set the ACLs.
+  - name: 'gcr.io/cloud-builders/gsutil'
+    args: ['cp', './echoprint-codegen', '${_ARTIFACT_URL}']
+
+  - name: 'gcr.io/cloud-builders/gsutil'
+    args: ['acl', 'ch', '-u', 'AllUsers:r', '${_ARTIFACT_URL}']

codegen.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+#
+# To make it easy to use with GNU Parallel, e.g.,
+#   parallel codegen.sh ::: *.mp3
+#
+echoprint-codegen -h "$1" > "$1.json"

examples/lookup.py

@@ -21,7 +21,7 @@ def lookup(file):
     fp = song.util.codegen(file)
     if len(fp) and "code" in fp[0]:
         # The version parameter to song/identify indicates the use of echoprint
-        result = song.identify(query_obj=fp, version="4.11")
+        result = song.identify(query_obj=fp, version="4.12")
         print "Got result:", result
         if len(result):
             print "Artist: %s (%s)" % (result[0].artist_name, result[0].artist_id)
@@ -36,4 +36,4 @@ def lookup(file):
     if len(sys.argv) < 2:
         print >>sys.stderr, "Usage: %s <audio file>" % sys.argv[0]
         sys.exit(1)
-    lookup(sys.argv[1])
+    lookup(sys.argv[1])

src/Codegen.cxx

@@ -23,6 +23,10 @@ using std::string;
 using std::vector;
 
 Codegen::Codegen(const float* pcm, unsigned int numSamples, int start_offset) {
+    for (int i = 0; i < 2; ++i) {
+        is_code_string_cached[i] = false;
+    }
+
     if (Params::AudioStreamInput::MaxSamples < (uint)numSamples)
         throw std::runtime_error("File was too big\n");
 
@@ -38,7 +42,11 @@ Codegen::Codegen(const float* pcm, unsigned int numSamples, int start_offset) {
     Fingerprint *pFingerprint = new Fingerprint(pSubbandAnalysis, start_offset);
     pFingerprint->Compute();
 
+#if defined(UNHASHED_CODES)
+    _CodeString = createCodeStringJSON(pFingerprint->getCodes());
+#else
     _CodeString = createCodeString(pFingerprint->getCodes());
+#endif
     _NumCodes = pFingerprint->getCodes().size();
 
     delete pFingerprint;
@@ -63,6 +71,24 @@ string Codegen::createCodeString(vector<FPCode> vCodes) {
     return compress(codestream.str());
 }
 
+string Codegen::createCodeStringJSON(vector<FPCode> vCodes) {
+    std::ostringstream codestream;
+    codestream << "[";
+    for (uint i = 0; i < vCodes.size(); i++) {
+        int hash = vCodes[i].code;
+        // codestream << std::setw(5) << hash;
+        codestream << "[" <<  vCodes[i].frame << ", "
+            << ((hash >> 20) & 7) << ", "
+            << ((hash >> 10) & 1023) << ", "
+            << ((hash >>  0) & 1023)
+            << "]";
+        if (i < vCodes.size()-1) {
+            codestream << ", ";
+        }
+    }
+    codestream << "]";
+    return codestream.str();
+}
 
 string Codegen::compress(const string& s) {
     long max_compressed_length = s.size()*2;
@@ -89,3 +115,24 @@ string Codegen::compress(const string& s) {
     delete [] compressed;
     return encoded;
 }
+
+std::string Codegen::getCodeString(bool human_readable) {
+    const uint n = human_readable;
+    if (!is_code_string_cached[n]) {
+        is_code_string_cached[n] = true;
+        if (human_readable) {
+            if (_CodeString.size() > 0) {
+                code_string_cache[n] = _CodeString;
+            } else {
+                code_string_cache[n] = "[]";
+            }
+        } else {
+            if (_CodeString.size() > 0) {
+                code_string_cache[n] = '"' + compress(_CodeString) + '"';
+            } else {
+                code_string_cache[n] = "\"\"";
+            }
+        }
+    }
+    return code_string_cache[n];
+}

src/Codegen.h

@@ -33,16 +33,20 @@ class CODEGEN_API Codegen {
 public:
     Codegen(const float* pcm, unsigned int numSamples, int start_offset);
 
-    std::string getCodeString(){return _CodeString;}
+    std::string getCodeString(bool human_readable);
+
     int getNumCodes(){return _NumCodes;}
     static double getVersion() { return ECHOPRINT_VERSION; }
 private:
     Fingerprint* computeFingerprint(SubbandAnalysis *pSubbandAnalysis, int start_offset);
     std::string createCodeString(std::vector<FPCode> vCodes);
+    std::string createCodeStringJSON(std::vector<FPCode> vCodes);
 
     std::string compress(const std::string& s);
     std::string _CodeString;
     int _NumCodes;
+    bool is_code_string_cached[2];
+    std::string code_string_cache[2];
 };
 
 #endif

src/Fingerprint.cxx

@@ -12,6 +12,8 @@
 #include "win_funcs.h"
 #endif
 
+#define SATURATE(var, val) if ((var) > (val)) (var) = (val);
+
 unsigned int MurmurHash2 ( const void * key, int len, unsigned int seed ) {
     // MurmurHash2, by Austin Appleby http://sites.google.com/site/murmurhash/
     // m and r are constants set by austin
@@ -182,13 +184,18 @@ uint Fingerprint::quantized_time_for_frame_absolute(uint frame) {
 
 void Fingerprint::Compute() {
     uint actual_codes = 0;
+#if !defined(UNHASHED_CODES)
     unsigned char hash_material[5];
     for(uint i=0;i<5;i++) hash_material[i] = 0;
+#endif
     uint * onset_counter_for_band;
     matrix_u out;
     uint onset_count = adaptiveOnsets(345, out, onset_counter_for_band);
     _Codes.resize(onset_count*6);
 
+#if defined(UNHASHED_CODES)
+    assert(SUBBANDS <= 8);
+#endif
     for(unsigned char band=0;band<SUBBANDS;band++) {
         if (onset_counter_for_band[band]>2) {
             for(uint onset=0;onset<onset_counter_for_band[band]-2;onset++) {
@@ -200,7 +207,7 @@ void Fingerprint::Compute() {
                     p[0][i] = 0;
                     p[1][i] = 0;
                 }
-                int nhashes = 6;
+                uint nhashes = 6;
 
                 if ((int)onset == (int)onset_counter_for_band[band]-4)  { nhashes = 3; }
                 if ((int)onset == (int)onset_counter_for_band[band]-3)  { nhashes = 1; }
@@ -222,16 +229,29 @@ void Fingerprint::Compute() {
                 }
 
                 // For each pair emit a code
-                for(uint k=0;k<6;k++) {
+                for(uint k=0; k < nhashes; k++) {
                     // Quantize the time deltas to 23ms
                     short time_delta0 = (short)quantized_time_for_frame_delta(p[0][k]);
                     short time_delta1 = (short)quantized_time_for_frame_delta(p[1][k]);
+                    if (k == 0 && time_delta0 == 0 && time_delta1 == 0) {
+                        continue;
+                    }
+                    uint hashed_code;
+#if defined(UNHASHED_CODES)
+                    assert(time_delta0 <= 1023);
+                    assert(time_delta1 <= 1023);
+#if defined(NDEBUG)
+                    SATURATE(time_delta0, 1023);
+                    SATURATE(time_delta1, 1023);
+#endif
+                    hashed_code = ((band & 7) << 20) | ((time_delta0 & 1023) << 10) | (time_delta1 & 1023);
+#else
                     // Create a key from the time deltas and the band index
                     memcpy(hash_material+0, (const void*)&time_delta0, 2);
                     memcpy(hash_material+2, (const void*)&time_delta1, 2);
                     memcpy(hash_material+4, (const void*)&band, 1);
-                    uint hashed_code = MurmurHash2(&hash_material, 5, HASH_SEED) & HASH_BITMASK;
-
+                    hashed_code = MurmurHash2(&hash_material, 5, HASH_SEED) & HASH_BITMASK;
+#endif
                     // Set the code alongside the time of onset
                     _Codes[actual_codes++] = FPCode(time_for_onset_ms_quantized, hashed_code);
                     //fprintf(stderr, "whee %d,%d: [%d, %d] (%d, %d), %d = %u at %d\n", actual_codes, k, time_delta0, time_delta1, p[0][k], p[1][k], band, hashed_code, time_for_onset_ms_quantized);

src/Makefile

@@ -1,18 +1,19 @@
 # Version of echoprint, as a list. Is expanded out
 # for various version numbers.
-EP_VERSION := 4 1 1
+EP_VERSION := 4 1 2
 VERSION := $(word 1, $(EP_VERSION)).$(word 2, $(EP_VERSION)).$(word 3, $(EP_VERSION))
 VERSION_MAJ := $(word 1, $(EP_VERSION))
 VERSION_COMPAT := $(word 1, $(EP_VERSION)).$(word 2, $(EP_VERSION))
 UNAME := $(shell uname -s)
 CXX=g++
 CC=gcc
-#OPTFLAGS=-g -O0
-OPTFLAGS=-O3 -DBOOST_UBLAS_NDEBUG -DNDEBUG
-BOOST_CFLAGS=-I/usr/local/include/boost-1_35
+#OPTFLAGS=-g -O0 -DUNHASHED_CODES
+OPTFLAGS=-O3 -DBOOST_UBLAS_NDEBUG -DNDEBUG -DUNHASHED_CODES
+BOOST_CFLAGS=-I/usr/local/include/boost
+JNI_CFLAGS=-I$$JAVA_HOME/include/ -I$$JAVA_HOME/include/darwin/ -I$$JAVA_HOME/include/linux/
 TAGLIB_CFLAGS=`taglib-config --cflags`
 TAGLIB_LIBS=`taglib-config --libs`
-CXXFLAGS=-Wall $(BOOST_CFLAGS) $(TAGLIB_CFLAGS) -fPIC $(OPTFLAGS)
+CXXFLAGS=-Wall $(BOOST_CFLAGS) $(TAGLIB_CFLAGS) $(JNI_CFLAGS) -fPIC $(OPTFLAGS)
 CFLAGS=-Wall -fPIC $(OPTFLAGS)
 LDFLAGS=$(TAGLIB_LIBS) -lz -lpthread $(OPTFLAGS)
 LIBNAME=libcodegen.so
@@ -26,7 +27,10 @@ MODULES_LIB = \
     Fingerprint.o \
     MatrixUtility.o \
     SubbandAnalysis.o \
-    Whitening.o
+    Whitening.o \
+	functions.o \
+	echoprint.o
+
 MODULES = $(MODULES_LIB) Metadata.o
 
 all: libcodegen echoprint-codegen
@@ -41,7 +45,7 @@ else
 endif
 
 echoprint-codegen: $(MODULES) main.o
-	$(CXX) $(MODULES) main.o $(LDFLAGS) -o ../echoprint-codegen
+	$(CXX) $(MODULES) main.o -o ../echoprint-codegen $(LDFLAGS)
 
 %.o: %.c %.h
 	$(CC) $(CFLAGS) -c -o $@ $<