cannam@3
|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
|
cannam@3
|
2
|
cannam@3
|
3 /*
|
cannam@3
|
4 Vamp
|
cannam@3
|
5
|
cannam@3
|
6 An API for audio analysis and feature extraction plugins.
|
cannam@3
|
7
|
cannam@3
|
8 Centre for Digital Music, Queen Mary, University of London.
|
cannam@3
|
9 Copyright 2006 Chris Cannam.
|
cannam@3
|
10
|
cannam@3
|
11 Permission is hereby granted, free of charge, to any person
|
cannam@3
|
12 obtaining a copy of this software and associated documentation
|
cannam@3
|
13 files (the "Software"), to deal in the Software without
|
cannam@3
|
14 restriction, including without limitation the rights to use, copy,
|
cannam@3
|
15 modify, merge, publish, distribute, sublicense, and/or sell copies
|
cannam@3
|
16 of the Software, and to permit persons to whom the Software is
|
cannam@3
|
17 furnished to do so, subject to the following conditions:
|
cannam@3
|
18
|
cannam@3
|
19 The above copyright notice and this permission notice shall be
|
cannam@3
|
20 included in all copies or substantial portions of the Software.
|
cannam@3
|
21
|
cannam@3
|
22 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
cannam@3
|
23 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
cannam@3
|
24 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
cannam@6
|
25 NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
|
cannam@3
|
26 ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
|
cannam@3
|
27 CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
|
cannam@3
|
28 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
cannam@3
|
29
|
cannam@3
|
30 Except as contained in this notice, the names of the Centre for
|
cannam@3
|
31 Digital Music; Queen Mary, University of London; and Chris Cannam
|
cannam@3
|
32 shall not be used in advertising or otherwise to promote the sale,
|
cannam@3
|
33 use or other dealings in this Software without prior written
|
cannam@3
|
34 authorization.
|
cannam@3
|
35 */
|
cannam@3
|
36
|
cannam@3
|
37 #ifndef _VAMP_PLUGIN_H_
|
cannam@3
|
38 #define _VAMP_PLUGIN_H_
|
cannam@3
|
39
|
cannam@3
|
40 #include "PluginBase.h"
|
cannam@3
|
41 #include "RealTime.h"
|
cannam@3
|
42
|
cannam@3
|
43 #include <string>
|
cannam@3
|
44 #include <vector>
|
cannam@3
|
45 #include <map>
|
cannam@3
|
46
|
cannam@3
|
47 namespace Vamp {
|
cannam@3
|
48
|
cannam@3
|
49 /**
|
cannam@3
|
50 * Vamp::Plugin is a base class for plugin instance classes
|
cannam@3
|
51 * that provide feature extraction from audio or related data.
|
cannam@3
|
52 *
|
cannam@3
|
53 * In most cases, the input will be audio and the output will be a
|
cannam@3
|
54 * stream of derived data at a lower sampling resolution than the
|
cannam@3
|
55 * input.
|
cannam@3
|
56 *
|
cannam@3
|
57 * Note that this class inherits several abstract methods from
|
cannam@53
|
58 * PluginBase. These must be implemented by the subclass.
|
cannam@53
|
59 *
|
cannam@53
|
60 *
|
cannam@53
|
61 * PLUGIN LIFECYCLE
|
cannam@3
|
62 *
|
cannam@3
|
63 * Feature extraction plugins are managed differently from real-time
|
cannam@3
|
64 * plugins (such as VST effects). The main difference is that the
|
cannam@3
|
65 * parameters for a feature extraction plugin are configured before
|
cannam@3
|
66 * the plugin is used, and do not change during use.
|
cannam@3
|
67 *
|
cannam@3
|
68 * 1. Host constructs the plugin, passing it the input sample rate.
|
cannam@3
|
69 * The plugin may do basic initialisation, but should not do anything
|
cannam@3
|
70 * computationally expensive at this point.
|
cannam@3
|
71 *
|
cannam@3
|
72 * 2. Host may query the plugin's available outputs.
|
cannam@3
|
73 *
|
cannam@3
|
74 * 3. Host queries programs and parameter descriptors, and may set
|
cannam@3
|
75 * some or all of them. Parameters that are not explicitly set should
|
cannam@3
|
76 * take their default values as specified in the parameter descriptor.
|
cannam@3
|
77 * When a program is set, the parameter values may change and the host
|
cannam@3
|
78 * will re-query them to check.
|
cannam@3
|
79 *
|
cannam@27
|
80 * 4. Host queries the preferred step size, block size and number of
|
cannam@27
|
81 * channels. These may all vary depending on the parameter values.
|
cannam@3
|
82 * (Note however that you cannot make the number of distinct outputs
|
cannam@27
|
83 * dependent on parameter values.)
|
cannam@3
|
84 *
|
cannam@3
|
85 * 5. Plugin is properly initialised with a call to initialise. This
|
cannam@3
|
86 * fixes the step size, block size, and number of channels, as well as
|
cannam@3
|
87 * all of the parameter and program settings. If the values passed in
|
cannam@3
|
88 * to initialise do not match the plugin's advertised preferred values
|
cannam@3
|
89 * from step 4, the plugin may refuse to initialise and return false
|
cannam@35
|
90 * (although if possible it should accept the new values). Any
|
cannam@35
|
91 * computationally expensive setup code should take place here.
|
cannam@3
|
92 *
|
cannam@27
|
93 * 6. Host finally checks the number of values per output (which may
|
cannam@27
|
94 * vary depending on the number of channels, step size and block size
|
cannam@27
|
95 * as well as the parameter values).
|
cannam@27
|
96 *
|
cannam@27
|
97 * 7. Host will repeatedly call the process method to pass in blocks
|
cannam@3
|
98 * of input data. This method may return features extracted from that
|
cannam@3
|
99 * data (if the plugin is causal).
|
cannam@3
|
100 *
|
cannam@27
|
101 * 8. Host will call getRemainingFeatures exactly once, after all the
|
cannam@3
|
102 * input data has been processed. This may return any non-causal or
|
cannam@3
|
103 * leftover features.
|
cannam@3
|
104 *
|
cannam@27
|
105 * 9. At any point after initialise was called, the host may
|
cannam@3
|
106 * optionally call the reset method and restart processing. (This
|
cannam@3
|
107 * does not mean it can change the parameters, which are fixed from
|
cannam@3
|
108 * initialise until destruction.)
|
cannam@3
|
109 *
|
cannam@3
|
110 * A plugin does not need to handle the case where setParameter or
|
cannam@3
|
111 * selectProgram is called after initialise has been called. It's the
|
cannam@35
|
112 * host's responsibility not to do that. Similarly, the plugin may
|
cannam@35
|
113 * safely assume that initialise is called no more than once.
|
cannam@3
|
114 */
|
cannam@3
|
115
|
cannam@3
|
116 class Plugin : public PluginBase
|
cannam@3
|
117 {
|
cannam@3
|
118 public:
|
cannam@20
|
119 virtual ~Plugin() { }
|
cannam@20
|
120
|
cannam@3
|
121 /**
|
cannam@3
|
122 * Initialise a plugin to prepare it for use with the given number
|
cannam@3
|
123 * of input channels, step size (window increment, in sample
|
cannam@3
|
124 * frames) and block size (window size, in sample frames).
|
cannam@3
|
125 *
|
cannam@3
|
126 * The input sample rate should have been already specified at
|
cannam@3
|
127 * construction time.
|
cannam@3
|
128 *
|
cannam@3
|
129 * Return true for successful initialisation, false if the number
|
cannam@3
|
130 * of input channels, step size and/or block size cannot be
|
cannam@3
|
131 * supported.
|
cannam@3
|
132 */
|
cannam@3
|
133 virtual bool initialise(size_t inputChannels,
|
cannam@3
|
134 size_t stepSize,
|
cannam@3
|
135 size_t blockSize) = 0;
|
cannam@3
|
136
|
cannam@3
|
137 /**
|
cannam@3
|
138 * Reset the plugin after use, to prepare it for another clean
|
cannam@3
|
139 * run. Not called for the first initialisation (i.e. initialise
|
cannam@3
|
140 * must also do a reset).
|
cannam@3
|
141 */
|
cannam@3
|
142 virtual void reset() = 0;
|
cannam@3
|
143
|
cannam@3
|
144 enum InputDomain { TimeDomain, FrequencyDomain };
|
cannam@3
|
145
|
cannam@3
|
146 /**
|
cannam@3
|
147 * Get the plugin's required input domain. If this is TimeDomain,
|
cannam@3
|
148 * the samples provided to the process() function (below) will be
|
cannam@3
|
149 * in the time domain, as for a traditional audio processing
|
cannam@3
|
150 * plugin. If this is FrequencyDomain, the host will carry out a
|
cannam@3
|
151 * windowed FFT of size equal to the negotiated block size on the
|
cannam@3
|
152 * data before passing the frequency bin data in to process().
|
cannam@3
|
153 * The plugin does not get to choose the window type -- the host
|
cannam@3
|
154 * will either let the user do so, or will use a Hanning window.
|
cannam@3
|
155 */
|
cannam@3
|
156 virtual InputDomain getInputDomain() const = 0;
|
cannam@3
|
157
|
cannam@3
|
158 /**
|
cannam@8
|
159 * Get the preferred block size (window size -- the number of
|
cannam@8
|
160 * sample frames passed in each block to the process() function).
|
cannam@8
|
161 * This should be called before initialise().
|
cannam@8
|
162 *
|
cannam@8
|
163 * A plugin that can handle any block size may return 0. The
|
cannam@8
|
164 * final block size will be set in the initialise() call.
|
cannam@8
|
165 */
|
cannam@8
|
166 virtual size_t getPreferredBlockSize() const { return 0; }
|
cannam@8
|
167
|
cannam@8
|
168 /**
|
cannam@3
|
169 * Get the preferred step size (window increment -- the distance
|
cannam@3
|
170 * in sample frames between the start frames of consecutive blocks
|
cannam@3
|
171 * passed to the process() function) for the plugin. This should
|
cannam@3
|
172 * be called before initialise().
|
cannam@8
|
173 *
|
cannam@8
|
174 * A plugin may return 0 if it has no particular interest in the
|
cannam@8
|
175 * step size. In this case, the host should make the step size
|
cannam@8
|
176 * equal to the block size if the plugin is accepting input in the
|
cannam@8
|
177 * time domain. If the plugin is accepting input in the frequency
|
cannam@8
|
178 * domain, the host may use any step size. The final step size
|
cannam@8
|
179 * will be set in the initialise() call.
|
cannam@3
|
180 */
|
cannam@8
|
181 virtual size_t getPreferredStepSize() const { return 0; }
|
cannam@3
|
182
|
cannam@3
|
183 /**
|
cannam@3
|
184 * Get the minimum supported number of input channels.
|
cannam@3
|
185 */
|
cannam@3
|
186 virtual size_t getMinChannelCount() const { return 1; }
|
cannam@3
|
187
|
cannam@3
|
188 /**
|
cannam@3
|
189 * Get the maximum supported number of input channels.
|
cannam@3
|
190 */
|
cannam@3
|
191 virtual size_t getMaxChannelCount() const { return 1; }
|
cannam@3
|
192
|
cannam@3
|
193 struct OutputDescriptor
|
cannam@3
|
194 {
|
cannam@3
|
195 /**
|
cannam@3
|
196 * The name of the output, in computer-usable form. Should be
|
cannam@49
|
197 * reasonably short and without whitespace or punctuation, using
|
cannam@49
|
198 * the characters [a-zA-Z0-9_] only.
|
cannam@49
|
199 * Example: "zero_crossing_count"
|
cannam@49
|
200 */
|
cannam@49
|
201 std::string identifier;
|
cannam@49
|
202
|
cannam@49
|
203 /**
|
cannam@49
|
204 * The human-readable name of the output.
|
cannam@49
|
205 * Example: "Zero Crossing Counts"
|
cannam@3
|
206 */
|
cannam@3
|
207 std::string name;
|
cannam@3
|
208
|
cannam@3
|
209 /**
|
cannam@49
|
210 * A human-readable short text describing the output. May be
|
cannam@49
|
211 * empty if the name has said it all already.
|
cannam@49
|
212 * Example: "The number of zero crossing points per processing block"
|
cannam@3
|
213 */
|
cannam@3
|
214 std::string description;
|
cannam@3
|
215
|
cannam@3
|
216 /**
|
cannam@3
|
217 * The unit of the output, in human-readable form.
|
cannam@3
|
218 */
|
cannam@3
|
219 std::string unit;
|
cannam@3
|
220
|
cannam@3
|
221 /**
|
cannam@9
|
222 * True if the output has the same number of values per sample
|
cannam@9
|
223 * for every output sample. Outputs for which this is false
|
cannam@3
|
224 * are unlikely to be very useful in a general-purpose host.
|
cannam@3
|
225 */
|
cannam@9
|
226 bool hasFixedBinCount;
|
cannam@3
|
227
|
cannam@3
|
228 /**
|
cannam@3
|
229 * The number of values per result of the output. Undefined
|
cannam@9
|
230 * if hasFixedBinCount is false. If this is zero, the output
|
cannam@3
|
231 * is point data (i.e. only the time of each output is of
|
cannam@3
|
232 * interest, the value list will be empty).
|
cannam@3
|
233 */
|
cannam@9
|
234 size_t binCount;
|
cannam@3
|
235
|
cannam@3
|
236 /**
|
cannam@49
|
237 * The (human-readable) names of each of the bins, if
|
cannam@49
|
238 * appropriate. This is always optional.
|
cannam@3
|
239 */
|
cannam@9
|
240 std::vector<std::string> binNames;
|
cannam@3
|
241
|
cannam@3
|
242 /**
|
cannam@9
|
243 * True if the results in each output bin fall within a fixed
|
cannam@9
|
244 * numeric range (minimum and maximum values). Undefined if
|
cannam@9
|
245 * binCount is zero.
|
cannam@3
|
246 */
|
cannam@3
|
247 bool hasKnownExtents;
|
cannam@3
|
248
|
cannam@3
|
249 /**
|
cannam@3
|
250 * Minimum value of the results in the output. Undefined if
|
cannam@9
|
251 * hasKnownExtents is false or binCount is zero.
|
cannam@3
|
252 */
|
cannam@3
|
253 float minValue;
|
cannam@3
|
254
|
cannam@3
|
255 /**
|
cannam@3
|
256 * Maximum value of the results in the output. Undefined if
|
cannam@9
|
257 * hasKnownExtents is false or binCount is zero.
|
cannam@3
|
258 */
|
cannam@3
|
259 float maxValue;
|
cannam@3
|
260
|
cannam@3
|
261 /**
|
cannam@3
|
262 * True if the output values are quantized to a particular
|
cannam@9
|
263 * resolution. Undefined if binCount is zero.
|
cannam@3
|
264 */
|
cannam@3
|
265 bool isQuantized;
|
cannam@3
|
266
|
cannam@3
|
267 /**
|
cannam@3
|
268 * Quantization resolution of the output values (e.g. 1.0 if
|
cannam@3
|
269 * they are all integers). Undefined if isQuantized is false
|
cannam@9
|
270 * or binCount is zero.
|
cannam@3
|
271 */
|
cannam@3
|
272 float quantizeStep;
|
cannam@3
|
273
|
cannam@3
|
274 enum SampleType {
|
cannam@3
|
275
|
cannam@3
|
276 /// Results from each process() align with that call's block start
|
cannam@3
|
277 OneSamplePerStep,
|
cannam@3
|
278
|
cannam@3
|
279 /// Results are evenly spaced in time (sampleRate specified below)
|
cannam@3
|
280 FixedSampleRate,
|
cannam@3
|
281
|
cannam@3
|
282 /// Results are unevenly spaced and have individual timestamps
|
cannam@3
|
283 VariableSampleRate
|
cannam@3
|
284 };
|
cannam@3
|
285
|
cannam@3
|
286 /**
|
cannam@3
|
287 * Positioning in time of the output results.
|
cannam@3
|
288 */
|
cannam@3
|
289 SampleType sampleType;
|
cannam@3
|
290
|
cannam@3
|
291 /**
|
cannam@17
|
292 * Sample rate of the output results, as samples per second.
|
cannam@17
|
293 * Undefined if sampleType is OneSamplePerStep.
|
cannam@3
|
294 *
|
cannam@3
|
295 * If sampleType is VariableSampleRate and this value is
|
cannam@3
|
296 * non-zero, then it may be used to calculate a resolution for
|
cannam@17
|
297 * the output (i.e. the "duration" of each sample, in time,
|
cannam@17
|
298 * will be 1/sampleRate seconds). It's recommended to set
|
cannam@17
|
299 * this to zero if that behaviour is not desired.
|
cannam@3
|
300 */
|
cannam@3
|
301 float sampleRate;
|
cannam@3
|
302 };
|
cannam@3
|
303
|
cannam@3
|
304 typedef std::vector<OutputDescriptor> OutputList;
|
cannam@3
|
305
|
cannam@3
|
306 /**
|
cannam@3
|
307 * Get the outputs of this plugin. An output's index in this list
|
cannam@3
|
308 * is used as its numeric index when looking it up in the
|
cannam@3
|
309 * FeatureSet returned from the process() call.
|
cannam@3
|
310 */
|
cannam@3
|
311 virtual OutputList getOutputDescriptors() const = 0;
|
cannam@3
|
312
|
cannam@3
|
313 struct Feature
|
cannam@3
|
314 {
|
cannam@3
|
315 /**
|
cannam@3
|
316 * True if an output feature has its own timestamp. This is
|
cannam@3
|
317 * mandatory if the output has VariableSampleRate, and is
|
cannam@3
|
318 * likely to be disregarded otherwise.
|
cannam@3
|
319 */
|
cannam@3
|
320 bool hasTimestamp;
|
cannam@3
|
321
|
cannam@3
|
322 /**
|
cannam@3
|
323 * Timestamp of the output feature. This is mandatory if the
|
cannam@3
|
324 * output has VariableSampleRate, and is likely to be
|
cannam@3
|
325 * disregarded otherwise. Undefined if hasTimestamp is false.
|
cannam@3
|
326 */
|
cannam@3
|
327 RealTime timestamp;
|
cannam@3
|
328
|
cannam@3
|
329 /**
|
cannam@3
|
330 * Results for a single sample of this feature. If the output
|
cannam@9
|
331 * hasFixedBinCount, there must be the same number of values
|
cannam@9
|
332 * as the output's binCount count.
|
cannam@3
|
333 */
|
cannam@3
|
334 std::vector<float> values;
|
cannam@3
|
335
|
cannam@3
|
336 /**
|
cannam@3
|
337 * Label for the sample of this feature.
|
cannam@3
|
338 */
|
cannam@3
|
339 std::string label;
|
cannam@3
|
340 };
|
cannam@3
|
341
|
cannam@3
|
342 typedef std::vector<Feature> FeatureList;
|
cannam@3
|
343 typedef std::map<int, FeatureList> FeatureSet; // key is output no
|
cannam@3
|
344
|
cannam@3
|
345 /**
|
cannam@3
|
346 * Process a single block of input data.
|
cannam@3
|
347 *
|
cannam@3
|
348 * If the plugin's inputDomain is TimeDomain, inputBuffers will
|
cannam@3
|
349 * point to one array of floats per input channel, and each of
|
cannam@3
|
350 * these arrays will contain blockSize consecutive audio samples
|
cannam@9
|
351 * (the host will zero-pad as necessary). The timestamp will be
|
cannam@9
|
352 * the real time in seconds of the start of the supplied block of
|
cannam@9
|
353 * samples.
|
cannam@3
|
354 *
|
cannam@3
|
355 * If the plugin's inputDomain is FrequencyDomain, inputBuffers
|
cannam@3
|
356 * will point to one array of floats per input channel, and each
|
cannam@47
|
357 * of these arrays will contain blockSize/2+1 consecutive pairs of
|
cannam@3
|
358 * real and imaginary component floats corresponding to bins
|
cannam@47
|
359 * 0..(blockSize/2) of the FFT output, where bin 0 contains the DC
|
cannam@47
|
360 * output and bin blockSize/2 corresponds to the Nyquist output.
|
cannam@61
|
361 * There will therefore be blockSize+2 floats per channel in total.
|
cannam@47
|
362 * The timestamp will be the real time in seconds of the centre of
|
cannam@47
|
363 * the FFT input window (i.e. the very first block passed to
|
cannam@47
|
364 * process might contain the FFT of half a block of zero samples
|
cannam@47
|
365 * and the first half-block of the actual data, with a timestamp
|
cannam@47
|
366 * of zero).
|
cannam@3
|
367 *
|
cannam@3
|
368 * Return any features that have become available after this
|
cannam@3
|
369 * process call. (These do not necessarily have to fall within
|
cannam@3
|
370 * the process block, except for OneSamplePerStep outputs.)
|
cannam@3
|
371 */
|
cannam@47
|
372 virtual FeatureSet process(const float *const *inputBuffers,
|
cannam@3
|
373 RealTime timestamp) = 0;
|
cannam@3
|
374
|
cannam@3
|
375 /**
|
cannam@3
|
376 * After all blocks have been processed, calculate and return any
|
cannam@3
|
377 * remaining features derived from the complete input.
|
cannam@3
|
378 */
|
cannam@3
|
379 virtual FeatureSet getRemainingFeatures() = 0;
|
cannam@3
|
380
|
cannam@53
|
381 /**
|
cannam@53
|
382 * Used to distinguish between Vamp::Plugin and other potential
|
cannam@59
|
383 * sibling subclasses of PluginBase. Do not reimplement this
|
cannam@53
|
384 * function in your subclass.
|
cannam@53
|
385 */
|
cannam@3
|
386 virtual std::string getType() const { return "Feature Extraction Plugin"; }
|
cannam@3
|
387
|
cannam@3
|
388 protected:
|
cannam@3
|
389 Plugin(float inputSampleRate) :
|
cannam@3
|
390 m_inputSampleRate(inputSampleRate) { }
|
cannam@3
|
391
|
cannam@3
|
392 float m_inputSampleRate;
|
cannam@3
|
393 };
|
cannam@3
|
394
|
cannam@3
|
395 }
|
cannam@3
|
396
|
cannam@3
|
397 #endif
|
cannam@3
|
398
|
cannam@3
|
399
|
cannam@3
|
400
|