libcm is a C development framework with an emphasis on audio signal processing applications.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

cmProc.h 33KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733
  1. #ifndef cmProc_h
  2. #define cmProc_h
  3. #ifdef __cplusplus
  4. extern "C" {
  5. #endif
  6. //( { file_desc:"Processor Library 1" kw:[proclib]}
  7. //)
  8. //( { label:cmAudioFileRd file_desc:"Audio file reader based on cmAudioFile" kw:[proc] }
  9. typedef struct
  10. {
  11. cmObj obj;
  12. cmAudioFileH_t h; // audio file handle
  13. cmAudioFileInfo_t info; // audio file info record
  14. unsigned chIdx;
  15. cmSample_t* outV; // buffer of audio from last read
  16. unsigned outN; // length of outV in samples
  17. cmChar_t* fn; // name of audio file
  18. unsigned lastReadFrmCnt; // count of samples actually read on last read
  19. bool eofFl;
  20. unsigned begFrmIdx;
  21. unsigned endFrmIdx;
  22. unsigned curFrmIdx; // frame index of the next frame to read
  23. cmMtxFile* mfp;
  24. } cmAudioFileRd;
  25. // set p to NULL to dynamically allocate the object
  26. // fn and chIdx are optional - set fn to NULL to allocate the reader without opening a file.
  27. // If fn is valid then chIdx must also be valid.
  28. // Set 'endSmpIdx' to cmInvalidIdx to return the entire signal in cmAudioFileRdRead().
  29. // Set 'endSmpIdx' to 0 to return all samples between 0 and the end of the file.
  30. cmAudioFileRd* cmAudioFileRdAlloc( cmCtx* c, cmAudioFileRd* p, unsigned procSmpCnt, const char* fn, unsigned chIdx, unsigned begSmpIdx, unsigned endSmpIdx );
  31. cmRC_t cmAudioFileRdFree( cmAudioFileRd** p );
  32. cmRC_t cmAudioFileRdOpen( cmAudioFileRd* p, unsigned procSmpCnt, const cmChar_t* fn, unsigned chIdx, unsigned begSmpIdx, unsigned endSmpIdx );
  33. cmRC_t cmAudioFileRdClose( cmAudioFileRd* p );
  34. // Returns cmEofRC if the end of file is encountered.
  35. cmRC_t cmAudioFileRdRead( cmAudioFileRd* p );
  36. cmRC_t cmAudioFileRdSeek( cmAudioFileRd* p, unsigned frmIdx );
  37. // Find the overall minimum, maximum, and mean sample values without changing the current file location.
  38. cmRC_t cmAudioFileRdMinMaxMean( cmAudioFileRd* p, unsigned chIdx, cmSample_t* minPtr, cmSample_t* maxPtr, cmSample_t* meanPtr );
  39. //)
  40. //( { label:cmShiftBuf file_desc:"Audio shift buffer processor" kw:[proc] }
  41. //------------------------------------------------------------------------------------------------------------
  42. // The buffer is intended to synchronize sample block rates between processes and to provide an overlapped
  43. // input buffer.
  44. typedef struct cmShiftBuf_str
  45. {
  46. cmObj obj;
  47. unsigned bufSmpCnt; // wndSmpCnt + hopSmpCnt
  48. cmSample_t* bufV; // bufV[bufSmpCnt] all other pointers use this memory
  49. cmSample_t* outV; // output window outV[ outN ]
  50. unsigned outN; // outN == wndSmpCnt
  51. unsigned procSmpCnt; // input sample count
  52. unsigned wndSmpCnt; // output sample count
  53. unsigned hopSmpCnt; // count of samples to shift the buffer by on each call to cmShiftExec()
  54. cmSample_t* inPtr; // ptr to location in outV[] to recv next sample
  55. bool fl; // reflects the last value returned by cmShiftBufExec().
  56. } cmShiftBuf;
  57. // Set p to NULL to dynamically allocate the object. hopSmpCnt must be <= wndSmpCnt.
  58. cmShiftBuf* cmShiftBufAlloc( cmCtx* c, cmShiftBuf* p, unsigned procSmpCnt, unsigned wndSmpCnt, unsigned hopSmpCnt );
  59. cmRC_t cmShiftBufFree( cmShiftBuf** p );
  60. cmRC_t cmShiftBufInit( cmShiftBuf* p, unsigned procSmpCnt, unsigned wndSmpCnt, unsigned hopSmpCnt );
  61. cmRC_t cmShiftBufFinal( cmShiftBuf* p );
  62. // Returns true if a new hop is ready to be read otherwise returns false.
  63. // In general cmShiftBufExec() should be called in a loop until it returns false.
  64. // Note that 'sp' and 'sn' are ignored except for the first call after the function returns false.
  65. // This means that when called in a loop 'sp' and 'sn' are only used on the first time through the loop.
  66. // When procSmpCnt is less than hopSmpCnt the loop will only execute when at least wndSmpCnt
  67. // new samples have been buffered.
  68. // When procSmpCnt is greater than hopSmpCnt the loop will execute multiple times until less
  69. // than wndSmpCnt new samples are available.
  70. // Note that 'sn' must always be less than or equal to procSmpCnt.
  71. //
  72. // Example:
  73. // while( fill(sp,sn) ) // fill sp[] with sn samples
  74. // {
  75. // // shift by hopSmpCnt samples on all passes - insert new samples on first pass
  76. // while( cmShiftBufExec(p,sp,sn) )
  77. // proc(p->outV,p->outN); // process p->outV[wndSmpCnt]
  78. // }
  79. bool cmShiftBufExec( cmShiftBuf* p, const cmSample_t* sp, unsigned sn );
  80. void cmShiftBufTest( cmCtx* c );
  81. //------------------------------------------------------------------------------------------------------------
  82. //)
  83. //( { label:cmWindowFunc file_desc:"Fourier Transform window function generator." kw:[proc]}
  84. enum
  85. {
  86. kInvalidWndId = 0x000,
  87. kHannWndId = 0x001,
  88. kHammingWndId = 0x002,
  89. kTriangleWndId = 0x004,
  90. kKaiserWndId = 0x008,
  91. kHannMatlabWndId= 0x010,
  92. kUnityWndId = 0x020,
  93. kWndIdMask = 0x0ff,
  94. kNormByLengthWndFl = 0x100, // mult by 1/wndSmpCnt
  95. kNormBySumWndFl = 0x200, // mult by wndSmpCnt/sum(wndV)
  96. kSlRejIsBetaWndFl = 0x400 // kaiserSideLobeRejectDb param. is actually kaiser beta arg.
  97. };
  98. typedef struct
  99. {
  100. cmObj obj;
  101. unsigned wndId;
  102. unsigned flags;
  103. cmSample_t* wndV;
  104. cmSample_t* outV;
  105. unsigned outN; // same as wndSmpCnt
  106. double kslRejectDb;
  107. cmMtxFile* mfp;
  108. } cmWndFunc;
  109. // Set p to NULL to dynamically allocate the object
  110. // if wndId is set to a valid value this function will internally call cmWndFuncInit()
  111. cmWndFunc* cmWndFuncAlloc( cmCtx* c, cmWndFunc* p, unsigned wndId, unsigned wndSmpCnt, double kaierSideLobeRejectDb );
  112. cmRC_t cmWndFuncFree( cmWndFunc** pp );
  113. cmRC_t cmWndFuncInit( cmWndFunc* p, unsigned wndId, unsigned wndSmpCnt, double kaiserSideLobeRejectDb );
  114. cmRC_t cmWndFuncFinal( cmWndFunc* p );
  115. cmRC_t cmWndFuncExec( cmWndFunc* p, const cmSample_t* sp, unsigned sn );
  116. void cmWndFuncTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  117. //------------------------------------------------------------------------------------------------------------
  118. //)
  119. //( { label:cmSpecDelay file_desc:"Spectral frame delay. A circular buffer for spectral (or other fixed length) vectors." kw:[proc]}
  120. typedef struct
  121. {
  122. cmObj obj;
  123. cmSample_t* bufPtr;
  124. unsigned maxDelayCnt;
  125. int inIdx;
  126. unsigned outN; // outN == binCnt
  127. } cmSpecDelay;
  128. // Set p to NULL to dynamically allocate the object.
  129. // Allocate a spectral frame delay capable of delaying for 'maxDelayCnt' hops and
  130. // where each vector contains 'binCnt' elements.
  131. cmSpecDelay* cmSpecDelayAlloc( cmCtx* c, cmSpecDelay* p, unsigned maxDelayCnt, unsigned binCnt );
  132. cmRC_t cmSpecDelayFree( cmSpecDelay** p );
  133. cmRC_t cmSpecDelayInit( cmSpecDelay* p, unsigned maxDelayCnt, unsigned binCnt );
  134. cmRC_t cmSpecDelayFinal(cmSpecDelay* p );
  135. // Give an input vector to the delay. 'sn' must <= binCnt
  136. cmRC_t cmSpecDelayExec( cmSpecDelay* p, const cmSample_t* sp, unsigned sn );
  137. // Get a pointer to a delayed vector. 'delayCnt' indicates the length of the delay in hops.
  138. // (e.g. 1 is the previous hop, 2 is two hops previous, ... )
  139. const cmSample_t* cmSpecDelayOutPtr(cmSpecDelay* p, unsigned delayCnt );
  140. //------------------------------------------------------------------------------------------------------------
  141. //)
  142. //( { label:cmFilter file_desc:"General purpose, LTI, Octave compatible, filter." kw:[proc] }
  143. typedef struct cmFilter_str
  144. {
  145. cmObj obj;
  146. cmReal_t* a; // feedback coeff's
  147. int an; // count of fb coeff's
  148. cmReal_t* b; // feedforward coeff's
  149. int bn; // count of ff coeffs'
  150. cmReal_t* d; // delay
  151. int di; //
  152. int cn; // length of delay
  153. cmReal_t b0; // 1st feedforward coeff
  154. cmSample_t* outSmpV; // signal output vector
  155. cmReal_t* outRealV;
  156. unsigned outN; // length of outV (procSmpCnt)
  157. } cmFilter;
  158. // d[dn] is the initial value of the delay line where dn = max(an,bn)-1.
  159. // Set d to NULL to intialize the delays to 0.
  160. cmFilter* cmFilterAlloc( cmCtx* c, cmFilter* p, const cmReal_t* b, unsigned bn, const cmReal_t* a, unsigned an, unsigned procSmpCnt, const cmReal_t* d );
  161. cmFilter* cmFilterAllocEllip( cmCtx* c, cmFilter* p, cmReal_t srate, cmReal_t passHz, cmReal_t stopHz, cmReal_t passDb, cmReal_t stopDb, unsigned procSmpCnt, const cmReal_t* d );
  162. cmRC_t cmFilterFree( cmFilter** pp );
  163. cmRC_t cmFilterInit( cmFilter* p, const cmReal_t* b, unsigned bn, const cmReal_t* a, unsigned an, unsigned procSmpCnt, const cmReal_t* d );
  164. cmRC_t cmFilterInitEllip( cmFilter* p, cmReal_t srate, cmReal_t passHz, cmReal_t stopHz, cmReal_t passDb, cmReal_t stopDb, unsigned procSmpCnt, const cmReal_t* d );
  165. cmRC_t cmFilterFinal( cmFilter* p );
  166. // If y==NULL or yn==0 then the output is sent to p->outV[p->outN].
  167. // This function can safely filter a signal in plcme therefore it is allowable for x[] and y[] to refer to the same memory.
  168. // If x[] overlaps y[] then y must be <= x.
  169. cmRC_t cmFilterExecS( cmFilter* p, const cmSample_t* x, unsigned xn, cmSample_t* y, unsigned yn );
  170. cmRC_t cmFilterExecR( cmFilter* p, const cmReal_t* x, unsigned xn, cmReal_t* y, unsigned yn );
  171. cmRC_t cmFilterSignal( cmCtx* c, const cmReal_t b[], unsigned bn, const cmReal_t a[], unsigned an, const cmSample_t* x, unsigned xn, cmSample_t* y, unsigned yn );
  172. // Perform forward-reverse filtering.
  173. cmRC_t cmFilterFilterS(cmCtx* c, const cmReal_t bb[], unsigned bn, const cmReal_t aa[], unsigned an, const cmSample_t* x, unsigned xn, cmSample_t* y, unsigned yn );
  174. cmRC_t cmFilterFilterR(cmCtx* c, const cmReal_t bb[], unsigned bn, const cmReal_t aa[], unsigned an, const cmReal_t* x, unsigned xn, cmReal_t* y, unsigned yn );
  175. void cmFilterTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  176. void cmFilterFilterTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  177. //------------------------------------------------------------------------------------------------------------
  178. //)
  179. //( { label:cmComplexDetect file_desc:"Complex domain onset detection function." kw:[proc] }
  180. typedef struct
  181. {
  182. cmObj obj;
  183. cmSpecDelay phsDelay;
  184. cmSpecDelay magDelay;
  185. unsigned binCnt;
  186. cmSample_t out;
  187. //cmMtxFile* mfp;
  188. //unsigned cdfSpRegId;
  189. } cmComplexDetect;
  190. // Set p to NULL to dynamically allocate the object.
  191. cmComplexDetect* cmComplexDetectAlloc(cmCtx* c, cmComplexDetect* p, unsigned binCnt );
  192. cmRC_t cmComplexDetectFree( cmComplexDetect** pp);
  193. cmRC_t cmComplexDetectInit( cmComplexDetect* p, unsigned binCnt );
  194. cmRC_t cmComplexDetectFinal(cmComplexDetect* p);
  195. cmRC_t cmComplexDetectExec( cmComplexDetect* p, const cmSample_t* magV, const cmSample_t* phsV, unsigned binCnt );
  196. //------------------------------------------------------------------------------------------------------------
  197. //)
  198. //( { label:cmComplexOnset file_desc:"Complex onset detection function" kw:[proc]}
  199. typedef struct
  200. {
  201. cmObj obj;
  202. double threshold;
  203. unsigned medSmpCnt;
  204. unsigned frmCnt; // expected number of frames to store
  205. unsigned dfi;
  206. cmSample_t* df;
  207. cmSample_t* fdf;
  208. cmSample_t onrate;
  209. //cmMtxFile* mfp;
  210. } cmComplexOnset;
  211. cmComplexOnset* cmComplexOnsetAlloc( cmCtx* c, cmComplexOnset* p, unsigned procSmpCnt, double srate, unsigned medFiltWndSmpCnt, double threshold, unsigned frameCnt );
  212. cmRC_t cmComplexOnsetFree( cmComplexOnset** pp);
  213. cmRC_t cmComplexOnsetInit( cmComplexOnset* p, unsigned procSmpCnt, double srate, unsigned medFiltWndSmpCnt, double threshold, unsigned frameCnt );
  214. cmRC_t cmComplexOnsetFinal( cmComplexOnset* p);
  215. cmRC_t cmComplexOnsetExec( cmComplexOnset* p, cmSample_t cdf );
  216. cmRC_t cmComplexOnsetCalc( cmComplexOnset* p );
  217. //------------------------------------------------------------------------------------------------------------
  218. //)
  219. //( { label:cmMfcc file_desc:"Mel Frequency Cepstral Coefficient (MFCC) measurement function." kw:[proc] }
  220. typedef struct
  221. {
  222. cmObj obj;
  223. unsigned melBandCnt;
  224. unsigned dctCoeffCnt;
  225. unsigned binCnt;
  226. cmReal_t* melM;
  227. cmReal_t* dctM;
  228. cmReal_t* outV;
  229. unsigned outN; // outN == dctCoeffCnt
  230. cmMtxFile* mfp;
  231. unsigned mfccSpRegId; // cmStatsProc regId
  232. } cmMfcc;
  233. cmMfcc* cmMfccAlloc( cmCtx* c, cmMfcc* p, double srate, unsigned melBandCnt, unsigned dctCoeffCnt, unsigned binCnt );
  234. cmRC_t cmMfccFree( cmMfcc** pp );
  235. cmRC_t cmMfccInit( cmMfcc* p, double srate, unsigned melBandCnt, unsigned dctCoeffCnt, unsigned binCnt );
  236. cmRC_t cmMfccFinal( cmMfcc* p );
  237. cmRC_t cmMfccExecPower( cmMfcc* p, const cmReal_t* magPowV, unsigned binCnt );
  238. cmRC_t cmMfccExecAmplitude( cmMfcc* p, const cmReal_t* magAmpV, unsigned binCnt );
  239. void cmMfccTest();
  240. //------------------------------------------------------------------------------------------------------------
  241. //)
  242. //( { label:cmSones file_desc:"Sones measurement function." kw:[proc] }
  243. typedef struct
  244. {
  245. cmObj obj;
  246. cmReal_t* ttmV; // Terhardt outer ear filter
  247. cmReal_t* sfM; // Shroeder spreading function
  248. unsigned* barkIdxV; // Bark to bin map
  249. unsigned* barkCntV; //
  250. cmReal_t* outV; // specific loudness in sones
  251. unsigned outN; // outN == barkBandCnt;
  252. cmReal_t overallLoudness; // overall loudness in sones
  253. unsigned binCnt; // expected length of incoming power spectrum
  254. unsigned barkBandCnt; // count of bark bands
  255. unsigned flags; //
  256. cmMtxFile* mfp;
  257. unsigned sonesSpRegId;
  258. unsigned loudSpRegId;
  259. } cmSones;
  260. enum { kDontUseEqlLoudSonesFl=0x00, kUseEqlLoudSonesFl=0x01 };
  261. cmSones* cmSonesAlloc( cmCtx* c, cmSones* p, double srate, unsigned barkBandCnt, unsigned binCnt, unsigned flags );
  262. cmRC_t cmSonesFree( cmSones** pp );
  263. cmRC_t cmSonesInit( cmSones* p, double srate, unsigned barkBandCnt, unsigned binCnt, unsigned flags );
  264. cmRC_t cmSonesFinal( cmSones* p );
  265. cmRC_t cmSonesExec( cmSones* p, const cmReal_t* magPowV, unsigned binCnt );
  266. void cmSonesTest();
  267. //------------------------------------------------------------------------------------------------------------
  268. //)
  269. //( { label: cmAudioOffsetScale file_desc:"Audio signal pre-processing normalizer." kw:[proc] }
  270. typedef struct
  271. {
  272. cmObj obj;
  273. unsigned cBufCnt;
  274. unsigned cBufCurCnt;
  275. unsigned cBufIdx;
  276. double cBufSum;
  277. unsigned cCntSum;
  278. cmReal_t* cBufPtr;
  279. unsigned* cCntPtr;
  280. cmSample_t offset;
  281. double dBref;
  282. cmSample_t* outV;
  283. unsigned outN; // (outN == procSmpCnt)
  284. unsigned flags;
  285. cmMtxFile* mfp;
  286. } cmAudioOffsetScale;
  287. // This processor adds an offset to an audio signal and scales into dB (SPL) using one of two techniques
  288. // 1) Measures the effective sound pressure (via RMS) and then scales the signal to the reference dB (SPL)
  289. // In this case dBref is commonly set to 70. See Timony, 2004, Implementing Loudness Models in Matlab.
  290. //
  291. // 2) treats the dBref as the maximum dB (SPL) and scales the signal by this amount without regard
  292. // measured signal level. In this case dBref is commonly set to 96 (max. dB (SPL) value for 16 bits)
  293. // and rmsWndSecs is ignored.
  294. //
  295. // Note that setting rmsWndSecs to zero has the effect of using procSmpCnt as the window length.
  296. enum { kNoAudioScaleFl=0x01, kRmsAudioScaleFl=0x02, kFixedAudioScaleFl=0x04 };
  297. cmAudioOffsetScale* cmAudioOffsetScaleAlloc( cmCtx* c, cmAudioOffsetScale* p, unsigned procSmpCnt, double srate, cmSample_t offset, double rmsWndSecs, double dBref, unsigned flags );
  298. cmRC_t cmAudioOffsetScaleFree( cmAudioOffsetScale** pp );
  299. cmRC_t cmAudioOffsetScaleInit( cmAudioOffsetScale* p, unsigned procSmpCnt, double srate, cmSample_t offset, double rmsWndSecs, double dBref, unsigned flags );
  300. cmRC_t cmAudioOffsetScaleFinal( cmAudioOffsetScale* p );
  301. cmRC_t cmAudioOffsetScaleExec( cmAudioOffsetScale* p, const cmSample_t* sp, unsigned sn );
  302. //------------------------------------------------------------------------------------------------------------
  303. //)
  304. //( { label:cmSpecMeas file_desc:"Measure a signals RMS, High-Frequency Content, Spectral Centroid, and Spectral Spread." kw:[proc]}
  305. typedef struct
  306. {
  307. cmObj obj;
  308. cmReal_t* rmsV;
  309. cmReal_t* hfcV;
  310. cmReal_t* scnV;
  311. cmReal_t rmsSum;
  312. cmReal_t hfcSum;
  313. cmReal_t scnSum;
  314. cmReal_t ssSum;
  315. cmReal_t rms; // RMS output
  316. cmReal_t hfc; // high-frequency content output
  317. cmReal_t sc; // spectral centroid output
  318. cmReal_t ss; // spectral spread output
  319. unsigned binCnt;
  320. unsigned flags;
  321. unsigned wndFrmCnt;
  322. unsigned frameIdx;
  323. unsigned frameCnt;
  324. double binHz;
  325. cmMtxFile* mfp;
  326. unsigned rmsSpRegId;
  327. unsigned hfcSpRegId;
  328. unsigned scSpRegId;
  329. unsigned ssSpRegId;
  330. } cmSpecMeas;
  331. // Set wndFrmCnt to the number of spectral frames to take the measurement over.
  332. // Setting wndFrmCnt to 1 has the effect of calculating the value on the current frame only.
  333. // Set flags = kWholeSigSpecMeasFl to ignore wndFrmCnt and calculate the result on the entire signal.
  334. // In effect this treats the entire signal as the length of the measurement window.
  335. enum { kWholeSigSpecMeasFl=0x00, kUseWndSpecMeasFl=0x01 };
  336. cmSpecMeas* cmSpecMeasAlloc( cmCtx* c, cmSpecMeas* p, double srate, unsigned binCnt, unsigned wndFrmCnt, unsigned flags );
  337. cmRC_t cmSpecMeasFree( cmSpecMeas** pp );
  338. cmRC_t cmSpecMeasInit( cmSpecMeas* p, double srate, unsigned binCnt, unsigned wndFrmCnt, unsigned flags );
  339. cmRC_t cmSpecMeasFinal( cmSpecMeas* p );
  340. cmRC_t cmSpecMeasExec( cmSpecMeas* p, const cmReal_t* magPowV, unsigned binCnt );
  341. //------------------------------------------------------------------------------------------------------------
  342. //)
  343. //( { label:cmSigMeas file_desc:"Measure a time domain signals zero crossing rate." kw:[proc]}
  344. typedef struct
  345. {
  346. cmObj obj;
  347. cmShiftBuf* sbp; // shift buffer used internally if procSmpCnt < measSmpCnt
  348. cmShiftBuf shiftBuf;
  349. double srate; //
  350. cmReal_t zcr; // zero crossing rate per second
  351. cmSample_t zcrDelay; // used internally by zero crossing count algorithm
  352. unsigned measSmpCnt; // length of measurement window in samples
  353. unsigned procSmpCnt; // expected number of samples per call to exec
  354. unsigned zcrSpRegId;
  355. cmMtxFile* mfp;
  356. } cmSigMeas;
  357. // procSmpCnt must be <= measSmpCnt
  358. cmSigMeas* cmSigMeasAlloc( cmCtx* c, cmSigMeas* p, double srate, unsigned procSmpCnt, unsigned measSmpCnt );
  359. cmRC_t cmSigMeasFree( cmSigMeas** pp );
  360. cmRC_t cmSigMeasInit( cmSigMeas* p, double srate, unsigned procSmpCnt, unsigned measSmpCnt );
  361. cmRC_t cmSigMeasFinal( cmSigMeas* p );
  362. cmRC_t cmSigMeasExec( cmSigMeas* p, const cmSample_t* sigV, unsigned smpCnt );
  363. //------------------------------------------------------------------------------------------------------------
  364. //)
  365. //( { label:cmSRC file_desc:"Sample rate converter" kw:[proc] }
  366. typedef struct
  367. {
  368. cmObj obj;
  369. cmFilter filt;
  370. cmSample_t* outV;
  371. unsigned outN;
  372. unsigned upFact;
  373. unsigned dnFact;
  374. unsigned upi;
  375. unsigned dni;
  376. cmMtxFile* mfp;
  377. } cmSRC;
  378. // The srate paramater is the sample rate of the source signal provided via cmSRCExec()
  379. cmSRC* cmSRCAlloc( cmCtx* c, cmSRC* p, double srate, unsigned procSmpCnt, unsigned upFact, unsigned dnFact );
  380. cmRC_t cmSRCFree( cmSRC** pp );
  381. cmRC_t cmSRCInit( cmSRC* p, double srate, unsigned procSmpCnt, unsigned upFact, unsigned dnFact );
  382. cmRC_t cmSRCFinal( cmSRC* p );
  383. cmRC_t cmSRCExec( cmSRC* p, const cmSample_t* sp, unsigned sn );
  384. void cmSRCTest();
  385. //------------------------------------------------------------------------------------------------------------
  386. //)
  387. //( { label:cmConstQ file_desc:"Contant-Q transform." kw:[proc] }
  388. typedef struct
  389. {
  390. cmObj obj;
  391. cmComplexR_t* fiV;
  392. cmComplexR_t* foV;
  393. cmComplexR_t* skM; // skM[ wndSmpCnt, constQBinCnt ]
  394. unsigned* skBegV; // skBegV[ constQBinCnt ] indexes used to decrease the size of the mtx mult in cmConstQExex()
  395. unsigned* skEndV; // skEndV[ constQBinCnt ]
  396. unsigned wndSmpCnt; // window length of the complex FFT required to feed this transform
  397. unsigned constQBinCnt; // count of bins in the const Q output
  398. unsigned binsPerOctave; //
  399. cmComplexR_t* outV; // outV[ constQBinCnt ]
  400. cmReal_t* magV; // outV[ constQBinCnt ]
  401. cmMtxFile* mfp;
  402. } cmConstQ;
  403. cmConstQ* cmConstQAlloc( cmCtx* c, cmConstQ* p, double srate, unsigned minMidiPitch, unsigned maxMidiPitch, unsigned binsPerOctave, double thresh );
  404. cmRC_t cmConstQFree( cmConstQ** pp );
  405. cmRC_t cmConstQInit( cmConstQ* p, double srate, unsigned minMidiPitch, unsigned maxMidiPitch, unsigned binsPerOctave, double thresh );
  406. cmRC_t cmConstQFinal( cmConstQ* p );
  407. cmRC_t cmConstQExec( cmConstQ* p, const cmComplexR_t* ftV, unsigned binCnt );
  408. //------------------------------------------------------------------------------------------------------------
  409. //)
  410. //( { label:cmTuneHpcp file_desc:"Generate a tuned chromagram." kw:[proc]}
  411. typedef struct
  412. {
  413. cmObj obj;
  414. cmReal_t* hpcpM; // hpcpM[ frameCnt , binsPerOctave ] - stored hpcp
  415. cmReal_t* fhpcpM;// fhpcpM[ binsPerOctave, frameCnt ] - filtered hpcp (note transposed relative to hpcpA)
  416. unsigned* histV; // histM[ binsPerOctave/12 ]
  417. cmReal_t* outM; // outM[ 12, frameCnt ];
  418. unsigned histN; // binsPerOctave/12
  419. unsigned binsPerOctave; // const-q bins representing 1 octave
  420. unsigned constQBinCnt; // total count of const-q bins
  421. unsigned frameCnt; // expected count of hpcp vectors to store.
  422. unsigned frameIdx; // next column in hpcpM[] to receive input
  423. unsigned cqMinMidiPitch;
  424. unsigned medFiltOrder;
  425. cmReal_t* meanV; // meanV[12]
  426. cmReal_t* varV; // varV[12]
  427. cmMtxFile* mf0p; // debug files
  428. cmMtxFile* mf1p;
  429. cmMtxFile* mf2p;
  430. } cmHpcp;
  431. cmHpcp* cmTunedHpcpAlloc( cmCtx* c, cmHpcp* p, unsigned binsPerOctave, unsigned constQBinCnt, unsigned cqMinMidiPitch, unsigned frameCnt, unsigned medFiltOrder );
  432. cmRC_t cmTunedHpcpFree( cmHpcp** pp );
  433. cmRC_t cmTunedHpcpInit( cmHpcp* p, unsigned binsPerOctave, unsigned constQBinCnt, unsigned cqMinMidiPitch, unsigned frameCnt, unsigned medFiltOrder );
  434. cmRC_t cmTunedHpcpFinal( cmHpcp* p );
  435. cmRC_t cmTunedHpcpExec( cmHpcp* p, const cmComplexR_t* constQBinPtr, unsigned constQBinCnt );
  436. cmRC_t cmTunedHpcpTuneAndFilter( cmHpcp* p);
  437. //------------------------------------------------------------------------------------------------------------
  438. //)
  439. //( { label:cmBeatHist file_desc:"Generate a beat candidate histogram." kw:[proc]}
  440. struct cmFftRR_str;
  441. struct cmIFftRR_str;
  442. typedef struct
  443. {
  444. cmObj obj;
  445. struct cmFftRR_str* fft;
  446. struct cmIFftRR_str* ifft;
  447. unsigned frmCnt; // 512 length of df
  448. unsigned maxLagCnt; // 128 length of longest CMF lag
  449. unsigned histBinCnt; // 15 count of histogram elements and rows in H[]
  450. unsigned hColCnt; // 128 count of columns in H[]
  451. cmReal_t* m; // m[ frmCnt x maxLagCnt ]
  452. cmReal_t* H; // histogram transformation mtx
  453. cmReal_t* df; // df[ frmCnt ] onset detection function
  454. cmReal_t* fdf; // fdf[ frmCnt ] filtered onset detection function
  455. unsigned dfi; // index next df[] location to receive an incoming value
  456. cmReal_t* histV; // histV[ histBinCnt ] histogram output
  457. cmMtxFile* mfp;
  458. } cmBeatHist;
  459. cmBeatHist* cmBeatHistAlloc( cmCtx* c, cmBeatHist* p, unsigned frmCnt );
  460. cmRC_t cmBeatHistFree( cmBeatHist** pp );
  461. cmRC_t cmBeatHistInit( cmBeatHist* p, unsigned frmCnt );
  462. cmRC_t cmBeatHistFinal( cmBeatHist* p );
  463. cmRC_t cmBeatHistExec( cmBeatHist* p, cmSample_t df );
  464. cmRC_t cmBeatHistCalc( cmBeatHist* p );
  465. //------------------------------------------------------------------------------------------------------------
  466. //)
  467. //( { label:cmGmm file_desc"Gaussian Mixture Model containing N Gaussian PDF's each of dimension D." kw:[proc model]}
  468. typedef struct
  469. {
  470. cmObj obj;
  471. unsigned K; // count of components
  472. unsigned D; // dimensionality of each component
  473. cmReal_t* gV; // gM[ K ] mixture gain vector
  474. cmReal_t* uM; // uM[ D x K ] component mean column vectors
  475. cmReal_t* sMM; // sMM[D x D x K ] component covariance matrices - each column is a DxD matrix
  476. cmReal_t* isMM; // isMM[D x D x K] inverted covar matrices
  477. cmReal_t* uMM; // uMM[ D x D x K] upper triangle factor of chol(sMM)
  478. cmReal_t* logDetV;// detV[ K ] determinent of covar matrices
  479. cmReal_t* t; // t[ D x D ]scratch matrix used for training
  480. unsigned uflags; // user defined flags
  481. } cmGmm_t;
  482. enum { cmMdgNoFlags=0x0, cmGmmDiagFl=0x01, cmGmmSkipKmeansFl=0x02 };
  483. cmGmm_t* cmGmmAlloc( cmCtx* c, cmGmm_t* p, unsigned N, unsigned D, const cmReal_t* gV, const cmReal_t* uM, const cmReal_t* sMM, unsigned flags );
  484. cmRC_t cmGmmFree( cmGmm_t** pp );
  485. cmRC_t cmGmmInit( cmGmm_t* p, unsigned N, unsigned D, const cmReal_t* gV, const cmReal_t* uM, const cmReal_t* sMM, unsigned flags );
  486. cmRC_t cmGmmFinal( cmGmm_t* p );
  487. // Estimate the parameters of the GMM using the training data in xM[p->D,xN].
  488. // *iterCntPtr on input is the number of iterations with no change in class assignment to signal convergence.
  489. // *iterCntPtr on output is the total number of interations required to converge.
  490. cmRC_t cmGmmTrain( cmGmm_t* p, const cmReal_t* xM, unsigned xN, unsigned* iterCntPtr );
  491. // Return a pointer to the feature vector at frmIdx containing D elements.
  492. typedef const cmReal_t* (*cmGmmReadFunc_t)( void* userPtr, unsigned colIdx );
  493. // Same as cmGmmTrain() but uses a function to access the feature vector.
  494. // The optional matrix uM[D,K] contains the initial mean values or NULL if not used.
  495. // The optional flag array roFlV[K] is used to indicate read-only components and is only used
  496. // when the uM[] arg. is non-NULL. Set roFlV[i] to true to indicate that the mean value supplied by
  497. // the uM[] arg. should not be alterned by the training process.
  498. // If 'maxIterCnt' is positive then it is the maximum number of iterations the training process will make
  499. // otherwise it is ignored.
  500. cmRC_t cmGmmTrain2( cmGmm_t* p, cmGmmReadFunc_t readFunc, void* userFuncPtr, unsigned xN, unsigned* iterCntPtr, const cmReal_t* uM, const bool* roFlV, int maxIterCnt );
  501. // Generate data yN data points from the GMM and store the result in yM[p->D,yN].
  502. cmRC_t cmGmmGenerate( cmGmm_t* p, cmReal_t* yM, unsigned yN );
  503. // Evaluate the probability of each column of xM[p->D,xN] and return the result in y[xN].
  504. // If yM[xN,K] is non-NULL then the individual component prob. values are returned
  505. cmRC_t cmGmmEval( cmGmm_t* p, const cmReal_t* xM, unsigned xN, cmReal_t* yV, cmReal_t* yM);
  506. // Same as cmGmmEval() but uses a a function to access each data vector
  507. cmRC_t cmGmmEval2( cmGmm_t* p, cmGmmReadFunc_t readFunc, void* userFuncPtr, unsigned xN, cmReal_t* yV, cmReal_t* yM);
  508. // Evaluate each component for a single data point
  509. // xV[D] - observed data point
  510. // yV[K] - output contains the evaluation for each component
  511. cmRC_t cmGmmEval3( cmGmm_t* p, const cmReal_t* xV, cmReal_t* yV );
  512. void cmGmmPrint( cmGmm_t* p, bool detailsFl );
  513. void cmGmmTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  514. //------------------------------------------------------------------------------------------------------------
  515. //)
  516. //( { label:cmChmm file_desc:"Continuous Hidden Markov Model" kw:[proc model]}
  517. typedef struct
  518. {
  519. cmObj obj;
  520. unsigned N; // count of states
  521. unsigned K; // count of components per mixture
  522. unsigned D; // dimensionality of the observation data
  523. cmReal_t* iV; // iV[ N ] initial state probability mtx
  524. cmReal_t* aM; // aM[ N x N] transition probability mtx
  525. cmGmm_t** bV; // bV[ N ] observation probability mtx (array of pointers to GMM's)
  526. cmReal_t* bM; // bM[ N,T] state-observation probability matrix
  527. cmMtxFile* mfp;
  528. } cmChmm_t;
  529. // Continuous HMM consisting of stateN states where the observations
  530. // associated with each state are generated by a Gaussian mixture PDF.
  531. // stateN - count of states
  532. // mixN - count of components in the mixtures
  533. // dimN - dimensionality of the observation data
  534. cmChmm_t* cmChmmAlloc( cmCtx* c, cmChmm_t* p, unsigned stateN, unsigned mixN, unsigned dimN, const cmReal_t* iV, const cmReal_t* aM );
  535. cmRC_t cmChmmFree( cmChmm_t** pp );
  536. cmRC_t cmChmmInit( cmChmm_t* p, unsigned stateN, unsigned mixN, unsigned dimN, const cmReal_t* iV, const cmReal_t* aM );
  537. cmRC_t cmChmmFinal( cmChmm_t* p );
  538. // Set the iV,aM and bV parameters to well-formed random values.
  539. cmRC_t cmChmmRandomize( cmChmm_t* p, const cmReal_t* oM, unsigned T );
  540. // Train the HMM using segmental k-means to initialize the model parameters.
  541. // threshProb is the min change in fit between the data and the model above which the procedure will continue to iterate.
  542. // maxIterCnt is the maximum number of iterations the algorithm will make without regard for threshProb.
  543. // iterCnt is the value of iterCnt used in the call cmChmmTrain() on each iteration
  544. cmRC_t cmChmmSegKMeans( cmChmm_t* p, const cmReal_t* oM, unsigned T, cmReal_t threshProb, unsigned maxIterCnt, unsigned iterCnt );
  545. cmRC_t cmChmmSetGmm( cmChmm_t* p, unsigned i, const cmReal_t* wV, const cmReal_t* uM, const cmReal_t* sMM, unsigned flags );
  546. // oM[D,T] - observation matrix
  547. // alphaM[N,T] - prob of being in each state and observtin oM(:,t)
  548. // logPrV[T] - (optional) record the log prob of the data given the model at each time step
  549. // Returns sum(logPrV[T])
  550. cmReal_t cmChmmForward( const cmChmm_t* p, const cmReal_t* oM, unsigned T, cmReal_t* alphaM, cmReal_t* logPrV );
  551. void cmChmmBackward( const cmChmm_t* p, const cmReal_t* oM, unsigned T, cmReal_t* betaM );
  552. // bM[N,T] the state-observation probability table is optional
  553. cmReal_t cmChmmCompare( const cmChmm_t* p0, const cmChmm_t* p1, unsigned T );
  554. // Generate a series of observations.
  555. // oM[ p->D , T ] - output matrix
  556. // sV[ T ] - optional vector to record the state used to generate the ith observation.
  557. cmRC_t cmChmmGenerate( const cmChmm_t* p, cmReal_t* oM, unsigned T, unsigned* sV );
  558. // Infer the HMM parameters (p->iV,p->aM,p->bV) from the observations oM[D,T]
  559. enum { kNoTrainMixCoeffChmmFl=0x01, kNoTrainMeanChmmFl=0x02, kNoTrainCovarChmmFl=0x04 };
  560. cmRC_t cmChmmTrain( cmChmm_t* p, const cmReal_t* oM, unsigned T, unsigned iterCnt, cmReal_t thresh, unsigned flags );
  561. // Determine the ML state sequence yV[T] given the observations oM[D,T].
  562. cmRC_t cmChmmDecode( cmChmm_t* p, const cmReal_t* oM, unsigned T, unsigned* yV );
  563. void cmChmmPrint( cmChmm_t* p );
  564. void cmChmmTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  565. //------------------------------------------------------------------------------------------------------------
  566. //)
  567. //( { label:cmChord file_desc:"HMM based chord recognizer." kw:[proc]}
  568. typedef struct
  569. {
  570. cmObj obj;
  571. cmChmm_t* h; // hmm
  572. unsigned N; // state count N=24
  573. unsigned D; // data dimension D=12
  574. unsigned S; // tonal space dim S=6
  575. unsigned T; // frames in chromaM
  576. cmReal_t* iV; // iV[N]
  577. cmReal_t* aM; // aM[N,N]
  578. cmReal_t* uM; // uM[D,N]
  579. cmReal_t* sMM; // sMM[D*D,N]
  580. cmReal_t* phiM; // phiM[S,T]
  581. cmReal_t* chromaM; // chromaM[D,T]
  582. cmReal_t* tsM; // tsM[S,T]
  583. cmReal_t* cdtsV; // cdts[1,T]
  584. cmReal_t triadSeqMode;
  585. cmReal_t triadSeqVar;
  586. cmReal_t triadIntMean;
  587. cmReal_t triadIntVar;
  588. cmReal_t* tsMeanV; // tsMeanV[S];
  589. cmReal_t* tsVarV; // tsVarV[S]
  590. cmReal_t cdtsMean;
  591. cmReal_t cdtsVar;
  592. } cmChord;
  593. cmChord* cmChordAlloc( cmCtx* c, cmChord* p, const cmReal_t* chromaM, unsigned T );
  594. cmRC_t cmChordFree( cmChord** p );
  595. cmRC_t cmChordInit( cmChord* p, const cmReal_t* chromaM, unsigned T );
  596. cmRC_t cmChordFinal( cmChord* p );
  597. void cmChordTest( cmRpt_t* rpt, cmLHeapH_t lhH, cmSymTblH_t stH );
  598. //------------------------------------------------------------------------------------------------------------
  599. //)
  600. #ifdef __cplusplus
  601. }
  602. #endif
  603. #endif