blob: 6ed23f43c131016397c66c7c22aea4a30f1ccd21 [file] [log] [blame]
niklase@google.com470e71d2011-07-07 08:21:25 +00001/*
andrew@webrtc.org648af742012-02-08 01:57:29 +00002 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
niklase@google.com470e71d2011-07-07 08:21:25 +00003 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +000011#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_INCLUDE_AUDIO_PROCESSING_H_
12#define WEBRTC_MODULES_AUDIO_PROCESSING_INCLUDE_AUDIO_PROCESSING_H_
niklase@google.com470e71d2011-07-07 08:21:25 +000013
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +000014#include <stddef.h> // size_t
henrikg@webrtc.org863b5362013-12-06 16:05:17 +000015#include <stdio.h> // FILE
ajm@google.com22e65152011-07-18 18:03:01 +000016
andrew@webrtc.org61e596f2013-07-25 18:28:29 +000017#include "webrtc/common.h"
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +000018#include "webrtc/modules/interface/module.h"
19#include "webrtc/typedefs.h"
niklase@google.com470e71d2011-07-07 08:21:25 +000020
bjornv@webrtc.org91d11b32013-03-05 16:53:09 +000021struct AecCore;
22
niklase@google.com470e71d2011-07-07 08:21:25 +000023namespace webrtc {
24
25class AudioFrame;
26class EchoCancellation;
27class EchoControlMobile;
28class GainControl;
29class HighPassFilter;
30class LevelEstimator;
31class NoiseSuppression;
32class VoiceDetection;
33
andrew@webrtc.org6b1e2192013-09-25 23:46:20 +000034// Use to enable the delay correction feature. This now engages an extended
35// filter mode in the AEC, along with robustness measures around the reported
36// system delays. It comes with a significant increase in AEC complexity, but is
37// much more robust to unreliable reported delays.
38//
39// Detailed changes to the algorithm:
40// - The filter length is changed from 48 to 128 ms. This comes with tuning of
41// several parameters: i) filter adaptation stepsize and error threshold;
42// ii) non-linear processing smoothing and overdrive.
43// - Option to ignore the reported delays on platforms which we deem
44// sufficiently unreliable. See WEBRTC_UNTRUSTED_DELAY in echo_cancellation.c.
45// - Faster startup times by removing the excessive "startup phase" processing
46// of reported delays.
47// - Much more conservative adjustments to the far-end read pointer. We smooth
48// the delay difference more heavily, and back off from the difference more.
49// Adjustments force a readaptation of the filter, so they should be avoided
50// except when really necessary.
51struct DelayCorrection {
52 DelayCorrection() : enabled(false) {}
53 DelayCorrection(bool enabled) : enabled(enabled) {}
54
55 bool enabled;
56};
57
niklase@google.com470e71d2011-07-07 08:21:25 +000058// The Audio Processing Module (APM) provides a collection of voice processing
59// components designed for real-time communications software.
60//
61// APM operates on two audio streams on a frame-by-frame basis. Frames of the
62// primary stream, on which all processing is applied, are passed to
63// |ProcessStream()|. Frames of the reverse direction stream, which are used for
64// analysis by some components, are passed to |AnalyzeReverseStream()|. On the
65// client-side, this will typically be the near-end (capture) and far-end
66// (render) streams, respectively. APM should be placed in the signal chain as
67// close to the audio hardware abstraction layer (HAL) as possible.
68//
69// On the server-side, the reverse stream will normally not be used, with
70// processing occurring on each incoming stream.
71//
72// Component interfaces follow a similar pattern and are accessed through
73// corresponding getters in APM. All components are disabled at create-time,
74// with default settings that are recommended for most situations. New settings
75// can be applied without enabling a component. Enabling a component triggers
76// memory allocation and initialization to allow it to start processing the
77// streams.
78//
79// Thread safety is provided with the following assumptions to reduce locking
80// overhead:
81// 1. The stream getters and setters are called from the same thread as
82// ProcessStream(). More precisely, stream functions are never called
83// concurrently with ProcessStream().
84// 2. Parameter getters are never called concurrently with the corresponding
85// setter.
86//
87// APM accepts only 16-bit linear PCM audio data in frames of 10 ms. Multiple
88// channels should be interleaved.
89//
90// Usage example, omitting error checking:
91// AudioProcessing* apm = AudioProcessing::Create(0);
niklase@google.com470e71d2011-07-07 08:21:25 +000092//
93// apm->high_pass_filter()->Enable(true);
94//
95// apm->echo_cancellation()->enable_drift_compensation(false);
96// apm->echo_cancellation()->Enable(true);
97//
98// apm->noise_reduction()->set_level(kHighSuppression);
99// apm->noise_reduction()->Enable(true);
100//
101// apm->gain_control()->set_analog_level_limits(0, 255);
102// apm->gain_control()->set_mode(kAdaptiveAnalog);
103// apm->gain_control()->Enable(true);
104//
105// apm->voice_detection()->Enable(true);
106//
107// // Start a voice call...
108//
109// // ... Render frame arrives bound for the audio HAL ...
110// apm->AnalyzeReverseStream(render_frame);
111//
112// // ... Capture frame arrives from the audio HAL ...
113// // Call required set_stream_ functions.
114// apm->set_stream_delay_ms(delay_ms);
115// apm->gain_control()->set_stream_analog_level(analog_level);
116//
117// apm->ProcessStream(capture_frame);
118//
119// // Call required stream_ functions.
120// analog_level = apm->gain_control()->stream_analog_level();
121// has_voice = apm->stream_has_voice();
122//
123// // Repeate render and capture processing for the duration of the call...
124// // Start a new call...
125// apm->Initialize();
126//
127// // Close the application...
andrew@webrtc.orgf3930e92013-09-18 22:37:32 +0000128// delete apm;
niklase@google.com470e71d2011-07-07 08:21:25 +0000129//
130class AudioProcessing : public Module {
131 public:
132 // Creates a APM instance, with identifier |id|. Use one instance for every
133 // primary audio stream requiring processing. On the client-side, this would
134 // typically be one instance for the near-end stream, and additional instances
135 // for each far-end stream which requires processing. On the server-side,
136 // this would typically be one instance for every incoming stream.
137 static AudioProcessing* Create(int id);
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000138 virtual ~AudioProcessing() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000139
niklase@google.com470e71d2011-07-07 08:21:25 +0000140 // Initializes internal states, while retaining all user settings. This
141 // should be called before beginning to process a new audio stream. However,
142 // it is not necessary to call before processing the first stream after
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000143 // creation. It is also not necessary to call if the audio parameters (sample
144 // rate and number of channels) have changed. Passing updated parameters
145 // directly to |ProcessStream()| and |AnalyzeReverseStream()| is permissible.
niklase@google.com470e71d2011-07-07 08:21:25 +0000146 virtual int Initialize() = 0;
147
andrew@webrtc.org61e596f2013-07-25 18:28:29 +0000148 // Pass down additional options which don't have explicit setters. This
149 // ensures the options are applied immediately.
150 virtual void SetExtraOptions(const Config& config) = 0;
151
aluebs@webrtc.org0b72f582013-11-19 15:17:51 +0000152 virtual int EnableExperimentalNs(bool enable) = 0;
153 virtual bool experimental_ns_enabled() const = 0;
154
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000155 // DEPRECATED: It is now possible to modify the sample rate directly in a call
156 // to |ProcessStream|.
niklase@google.com470e71d2011-07-07 08:21:25 +0000157 // Sets the sample |rate| in Hz for both the primary and reverse audio
158 // streams. 8000, 16000 or 32000 Hz are permitted.
159 virtual int set_sample_rate_hz(int rate) = 0;
160 virtual int sample_rate_hz() const = 0;
161
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000162 // DEPRECATED: It is now possible to modify the number of channels directly in
163 // a call to |ProcessStream|.
niklase@google.com470e71d2011-07-07 08:21:25 +0000164 // Sets the number of channels for the primary audio stream. Input frames must
165 // contain a number of channels given by |input_channels|, while output frames
166 // will be returned with number of channels given by |output_channels|.
167 virtual int set_num_channels(int input_channels, int output_channels) = 0;
168 virtual int num_input_channels() const = 0;
169 virtual int num_output_channels() const = 0;
170
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000171 // DEPRECATED: It is now possible to modify the number of channels directly in
172 // a call to |AnalyzeReverseStream|.
niklase@google.com470e71d2011-07-07 08:21:25 +0000173 // Sets the number of channels for the reverse audio stream. Input frames must
174 // contain a number of channels given by |channels|.
175 virtual int set_num_reverse_channels(int channels) = 0;
176 virtual int num_reverse_channels() const = 0;
177
178 // Processes a 10 ms |frame| of the primary audio stream. On the client-side,
179 // this is the near-end (or captured) audio.
180 //
181 // If needed for enabled functionality, any function with the set_stream_ tag
182 // must be called prior to processing the current frame. Any getter function
183 // with the stream_ tag which is needed should be called after processing.
184 //
andrew@webrtc.org63a50982012-05-02 23:56:37 +0000185 // The |sample_rate_hz_|, |num_channels_|, and |samples_per_channel_|
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000186 // members of |frame| must be valid. If changed from the previous call to this
187 // method, it will trigger an initialization.
niklase@google.com470e71d2011-07-07 08:21:25 +0000188 virtual int ProcessStream(AudioFrame* frame) = 0;
189
190 // Analyzes a 10 ms |frame| of the reverse direction audio stream. The frame
191 // will not be modified. On the client-side, this is the far-end (or to be
192 // rendered) audio.
193 //
194 // It is only necessary to provide this if echo processing is enabled, as the
195 // reverse stream forms the echo reference signal. It is recommended, but not
196 // necessary, to provide if gain control is enabled. On the server-side this
197 // typically will not be used. If you're not sure what to pass in here,
198 // chances are you don't need to use it.
199 //
andrew@webrtc.org63a50982012-05-02 23:56:37 +0000200 // The |sample_rate_hz_|, |num_channels_|, and |samples_per_channel_|
andrew@webrtc.org60730cf2014-01-07 17:45:09 +0000201 // members of |frame| must be valid. |sample_rate_hz_| must correspond to
202 // |sample_rate_hz()|
niklase@google.com470e71d2011-07-07 08:21:25 +0000203 //
204 // TODO(ajm): add const to input; requires an implementation fix.
205 virtual int AnalyzeReverseStream(AudioFrame* frame) = 0;
206
207 // This must be called if and only if echo processing is enabled.
208 //
209 // Sets the |delay| in ms between AnalyzeReverseStream() receiving a far-end
210 // frame and ProcessStream() receiving a near-end frame containing the
211 // corresponding echo. On the client-side this can be expressed as
212 // delay = (t_render - t_analyze) + (t_process - t_capture)
213 // where,
214 // - t_analyze is the time a frame is passed to AnalyzeReverseStream() and
215 // t_render is the time the first sample of the same frame is rendered by
216 // the audio hardware.
217 // - t_capture is the time the first sample of a frame is captured by the
218 // audio hardware and t_pull is the time the same frame is passed to
219 // ProcessStream().
220 virtual int set_stream_delay_ms(int delay) = 0;
221 virtual int stream_delay_ms() const = 0;
222
andrew@webrtc.org6f9f8172012-03-06 19:03:39 +0000223 // Sets a delay |offset| in ms to add to the values passed in through
224 // set_stream_delay_ms(). May be positive or negative.
225 //
226 // Note that this could cause an otherwise valid value passed to
227 // set_stream_delay_ms() to return an error.
228 virtual void set_delay_offset_ms(int offset) = 0;
229 virtual int delay_offset_ms() const = 0;
230
niklase@google.com470e71d2011-07-07 08:21:25 +0000231 // Starts recording debugging information to a file specified by |filename|,
232 // a NULL-terminated string. If there is an ongoing recording, the old file
233 // will be closed, and recording will continue in the newly specified file.
234 // An already existing file will be overwritten without warning.
andrew@webrtc.org5ae19de2011-12-13 22:59:33 +0000235 static const size_t kMaxFilenameSize = 1024;
niklase@google.com470e71d2011-07-07 08:21:25 +0000236 virtual int StartDebugRecording(const char filename[kMaxFilenameSize]) = 0;
237
henrikg@webrtc.org863b5362013-12-06 16:05:17 +0000238 // Same as above but uses an existing file handle. Takes ownership
239 // of |handle| and closes it at StopDebugRecording().
240 virtual int StartDebugRecording(FILE* handle) = 0;
241
niklase@google.com470e71d2011-07-07 08:21:25 +0000242 // Stops recording debugging information, and closes the file. Recording
243 // cannot be resumed in the same file (without overwriting it).
244 virtual int StopDebugRecording() = 0;
245
246 // These provide access to the component interfaces and should never return
247 // NULL. The pointers will be valid for the lifetime of the APM instance.
248 // The memory for these objects is entirely managed internally.
249 virtual EchoCancellation* echo_cancellation() const = 0;
250 virtual EchoControlMobile* echo_control_mobile() const = 0;
251 virtual GainControl* gain_control() const = 0;
252 virtual HighPassFilter* high_pass_filter() const = 0;
253 virtual LevelEstimator* level_estimator() const = 0;
254 virtual NoiseSuppression* noise_suppression() const = 0;
255 virtual VoiceDetection* voice_detection() const = 0;
256
257 struct Statistic {
258 int instant; // Instantaneous value.
259 int average; // Long-term average.
260 int maximum; // Long-term maximum.
261 int minimum; // Long-term minimum.
262 };
263
andrew@webrtc.org648af742012-02-08 01:57:29 +0000264 enum Error {
265 // Fatal errors.
niklase@google.com470e71d2011-07-07 08:21:25 +0000266 kNoError = 0,
267 kUnspecifiedError = -1,
268 kCreationFailedError = -2,
269 kUnsupportedComponentError = -3,
270 kUnsupportedFunctionError = -4,
271 kNullPointerError = -5,
272 kBadParameterError = -6,
273 kBadSampleRateError = -7,
274 kBadDataLengthError = -8,
275 kBadNumberChannelsError = -9,
276 kFileError = -10,
277 kStreamParameterNotSetError = -11,
andrew@webrtc.org648af742012-02-08 01:57:29 +0000278 kNotEnabledError = -12,
niklase@google.com470e71d2011-07-07 08:21:25 +0000279
andrew@webrtc.org648af742012-02-08 01:57:29 +0000280 // Warnings are non-fatal.
niklase@google.com470e71d2011-07-07 08:21:25 +0000281 // This results when a set_stream_ parameter is out of range. Processing
282 // will continue, but the parameter may have been truncated.
andrew@webrtc.org648af742012-02-08 01:57:29 +0000283 kBadStreamParameterWarning = -13
niklase@google.com470e71d2011-07-07 08:21:25 +0000284 };
285
286 // Inherited from Module.
pbos@webrtc.org91620802013-08-02 11:44:11 +0000287 virtual int32_t TimeUntilNextProcess() OVERRIDE;
288 virtual int32_t Process() OVERRIDE;
niklase@google.com470e71d2011-07-07 08:21:25 +0000289};
290
291// The acoustic echo cancellation (AEC) component provides better performance
292// than AECM but also requires more processing power and is dependent on delay
293// stability and reporting accuracy. As such it is well-suited and recommended
294// for PC and IP phone applications.
295//
296// Not recommended to be enabled on the server-side.
297class EchoCancellation {
298 public:
299 // EchoCancellation and EchoControlMobile may not be enabled simultaneously.
300 // Enabling one will disable the other.
301 virtual int Enable(bool enable) = 0;
302 virtual bool is_enabled() const = 0;
303
304 // Differences in clock speed on the primary and reverse streams can impact
305 // the AEC performance. On the client-side, this could be seen when different
306 // render and capture devices are used, particularly with webcams.
307 //
308 // This enables a compensation mechanism, and requires that
309 // |set_device_sample_rate_hz()| and |set_stream_drift_samples()| be called.
310 virtual int enable_drift_compensation(bool enable) = 0;
311 virtual bool is_drift_compensation_enabled() const = 0;
312
313 // Provides the sampling rate of the audio devices. It is assumed the render
314 // and capture devices use the same nominal sample rate. Required if and only
315 // if drift compensation is enabled.
316 virtual int set_device_sample_rate_hz(int rate) = 0;
317 virtual int device_sample_rate_hz() const = 0;
318
319 // Sets the difference between the number of samples rendered and captured by
320 // the audio devices since the last call to |ProcessStream()|. Must be called
andrew@webrtc.org6be1e932013-03-01 18:47:28 +0000321 // if drift compensation is enabled, prior to |ProcessStream()|.
322 virtual void set_stream_drift_samples(int drift) = 0;
niklase@google.com470e71d2011-07-07 08:21:25 +0000323 virtual int stream_drift_samples() const = 0;
324
325 enum SuppressionLevel {
326 kLowSuppression,
327 kModerateSuppression,
328 kHighSuppression
329 };
330
331 // Sets the aggressiveness of the suppressor. A higher level trades off
332 // double-talk performance for increased echo suppression.
333 virtual int set_suppression_level(SuppressionLevel level) = 0;
334 virtual SuppressionLevel suppression_level() const = 0;
335
336 // Returns false if the current frame almost certainly contains no echo
337 // and true if it _might_ contain echo.
338 virtual bool stream_has_echo() const = 0;
339
340 // Enables the computation of various echo metrics. These are obtained
341 // through |GetMetrics()|.
342 virtual int enable_metrics(bool enable) = 0;
343 virtual bool are_metrics_enabled() const = 0;
344
345 // Each statistic is reported in dB.
346 // P_far: Far-end (render) signal power.
347 // P_echo: Near-end (capture) echo signal power.
348 // P_out: Signal power at the output of the AEC.
349 // P_a: Internal signal power at the point before the AEC's non-linear
350 // processor.
351 struct Metrics {
352 // RERL = ERL + ERLE
353 AudioProcessing::Statistic residual_echo_return_loss;
354
355 // ERL = 10log_10(P_far / P_echo)
356 AudioProcessing::Statistic echo_return_loss;
357
358 // ERLE = 10log_10(P_echo / P_out)
359 AudioProcessing::Statistic echo_return_loss_enhancement;
360
361 // (Pre non-linear processing suppression) A_NLP = 10log_10(P_echo / P_a)
362 AudioProcessing::Statistic a_nlp;
363 };
364
365 // TODO(ajm): discuss the metrics update period.
366 virtual int GetMetrics(Metrics* metrics) = 0;
367
bjornv@google.com1ba3dbe2011-10-03 08:18:10 +0000368 // Enables computation and logging of delay values. Statistics are obtained
369 // through |GetDelayMetrics()|.
370 virtual int enable_delay_logging(bool enable) = 0;
371 virtual bool is_delay_logging_enabled() const = 0;
372
373 // The delay metrics consists of the delay |median| and the delay standard
374 // deviation |std|. The values are averaged over the time period since the
375 // last call to |GetDelayMetrics()|.
376 virtual int GetDelayMetrics(int* median, int* std) = 0;
377
bjornv@webrtc.org91d11b32013-03-05 16:53:09 +0000378 // Returns a pointer to the low level AEC component. In case of multiple
379 // channels, the pointer to the first one is returned. A NULL pointer is
380 // returned when the AEC component is disabled or has not been initialized
381 // successfully.
382 virtual struct AecCore* aec_core() const = 0;
383
niklase@google.com470e71d2011-07-07 08:21:25 +0000384 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000385 virtual ~EchoCancellation() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000386};
387
388// The acoustic echo control for mobile (AECM) component is a low complexity
389// robust option intended for use on mobile devices.
390//
391// Not recommended to be enabled on the server-side.
392class EchoControlMobile {
393 public:
394 // EchoCancellation and EchoControlMobile may not be enabled simultaneously.
395 // Enabling one will disable the other.
396 virtual int Enable(bool enable) = 0;
397 virtual bool is_enabled() const = 0;
398
399 // Recommended settings for particular audio routes. In general, the louder
400 // the echo is expected to be, the higher this value should be set. The
401 // preferred setting may vary from device to device.
402 enum RoutingMode {
403 kQuietEarpieceOrHeadset,
404 kEarpiece,
405 kLoudEarpiece,
406 kSpeakerphone,
407 kLoudSpeakerphone
408 };
409
410 // Sets echo control appropriate for the audio routing |mode| on the device.
411 // It can and should be updated during a call if the audio routing changes.
412 virtual int set_routing_mode(RoutingMode mode) = 0;
413 virtual RoutingMode routing_mode() const = 0;
414
415 // Comfort noise replaces suppressed background noise to maintain a
416 // consistent signal level.
417 virtual int enable_comfort_noise(bool enable) = 0;
418 virtual bool is_comfort_noise_enabled() const = 0;
419
bjornv@google.comc4b939c2011-07-13 08:09:56 +0000420 // A typical use case is to initialize the component with an echo path from a
ajm@google.com22e65152011-07-18 18:03:01 +0000421 // previous call. The echo path is retrieved using |GetEchoPath()|, typically
422 // at the end of a call. The data can then be stored for later use as an
423 // initializer before the next call, using |SetEchoPath()|.
424 //
bjornv@google.comc4b939c2011-07-13 08:09:56 +0000425 // Controlling the echo path this way requires the data |size_bytes| to match
426 // the internal echo path size. This size can be acquired using
427 // |echo_path_size_bytes()|. |SetEchoPath()| causes an entire reset, worth
ajm@google.com22e65152011-07-18 18:03:01 +0000428 // noting if it is to be called during an ongoing call.
429 //
430 // It is possible that version incompatibilities may result in a stored echo
431 // path of the incorrect size. In this case, the stored path should be
432 // discarded.
433 virtual int SetEchoPath(const void* echo_path, size_t size_bytes) = 0;
434 virtual int GetEchoPath(void* echo_path, size_t size_bytes) const = 0;
435
436 // The returned path size is guaranteed not to change for the lifetime of
437 // the application.
438 static size_t echo_path_size_bytes();
bjornv@google.comc4b939c2011-07-13 08:09:56 +0000439
niklase@google.com470e71d2011-07-07 08:21:25 +0000440 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000441 virtual ~EchoControlMobile() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000442};
443
444// The automatic gain control (AGC) component brings the signal to an
445// appropriate range. This is done by applying a digital gain directly and, in
446// the analog mode, prescribing an analog gain to be applied at the audio HAL.
447//
448// Recommended to be enabled on the client-side.
449class GainControl {
450 public:
451 virtual int Enable(bool enable) = 0;
452 virtual bool is_enabled() const = 0;
453
454 // When an analog mode is set, this must be called prior to |ProcessStream()|
455 // to pass the current analog level from the audio HAL. Must be within the
456 // range provided to |set_analog_level_limits()|.
457 virtual int set_stream_analog_level(int level) = 0;
458
459 // When an analog mode is set, this should be called after |ProcessStream()|
460 // to obtain the recommended new analog level for the audio HAL. It is the
461 // users responsibility to apply this level.
462 virtual int stream_analog_level() = 0;
463
464 enum Mode {
465 // Adaptive mode intended for use if an analog volume control is available
466 // on the capture device. It will require the user to provide coupling
467 // between the OS mixer controls and AGC through the |stream_analog_level()|
468 // functions.
469 //
470 // It consists of an analog gain prescription for the audio device and a
471 // digital compression stage.
472 kAdaptiveAnalog,
473
474 // Adaptive mode intended for situations in which an analog volume control
475 // is unavailable. It operates in a similar fashion to the adaptive analog
476 // mode, but with scaling instead applied in the digital domain. As with
477 // the analog mode, it additionally uses a digital compression stage.
478 kAdaptiveDigital,
479
480 // Fixed mode which enables only the digital compression stage also used by
481 // the two adaptive modes.
482 //
483 // It is distinguished from the adaptive modes by considering only a
484 // short time-window of the input signal. It applies a fixed gain through
485 // most of the input level range, and compresses (gradually reduces gain
486 // with increasing level) the input signal at higher levels. This mode is
487 // preferred on embedded devices where the capture signal level is
488 // predictable, so that a known gain can be applied.
489 kFixedDigital
490 };
491
492 virtual int set_mode(Mode mode) = 0;
493 virtual Mode mode() const = 0;
494
495 // Sets the target peak |level| (or envelope) of the AGC in dBFs (decibels
496 // from digital full-scale). The convention is to use positive values. For
497 // instance, passing in a value of 3 corresponds to -3 dBFs, or a target
498 // level 3 dB below full-scale. Limited to [0, 31].
499 //
500 // TODO(ajm): use a negative value here instead, if/when VoE will similarly
501 // update its interface.
502 virtual int set_target_level_dbfs(int level) = 0;
503 virtual int target_level_dbfs() const = 0;
504
505 // Sets the maximum |gain| the digital compression stage may apply, in dB. A
506 // higher number corresponds to greater compression, while a value of 0 will
507 // leave the signal uncompressed. Limited to [0, 90].
508 virtual int set_compression_gain_db(int gain) = 0;
509 virtual int compression_gain_db() const = 0;
510
511 // When enabled, the compression stage will hard limit the signal to the
512 // target level. Otherwise, the signal will be compressed but not limited
513 // above the target level.
514 virtual int enable_limiter(bool enable) = 0;
515 virtual bool is_limiter_enabled() const = 0;
516
517 // Sets the |minimum| and |maximum| analog levels of the audio capture device.
518 // Must be set if and only if an analog mode is used. Limited to [0, 65535].
519 virtual int set_analog_level_limits(int minimum,
520 int maximum) = 0;
521 virtual int analog_level_minimum() const = 0;
522 virtual int analog_level_maximum() const = 0;
523
524 // Returns true if the AGC has detected a saturation event (period where the
525 // signal reaches digital full-scale) in the current frame and the analog
526 // level cannot be reduced.
527 //
528 // This could be used as an indicator to reduce or disable analog mic gain at
529 // the audio HAL.
530 virtual bool stream_is_saturated() const = 0;
531
532 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000533 virtual ~GainControl() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000534};
535
536// A filtering component which removes DC offset and low-frequency noise.
537// Recommended to be enabled on the client-side.
538class HighPassFilter {
539 public:
540 virtual int Enable(bool enable) = 0;
541 virtual bool is_enabled() const = 0;
542
543 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000544 virtual ~HighPassFilter() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000545};
546
547// An estimation component used to retrieve level metrics.
548class LevelEstimator {
549 public:
550 virtual int Enable(bool enable) = 0;
551 virtual bool is_enabled() const = 0;
552
andrew@webrtc.org755b04a2011-11-15 16:57:56 +0000553 // Returns the root mean square (RMS) level in dBFs (decibels from digital
554 // full-scale), or alternately dBov. It is computed over all primary stream
555 // frames since the last call to RMS(). The returned value is positive but
556 // should be interpreted as negative. It is constrained to [0, 127].
557 //
558 // The computation follows:
559 // http://tools.ietf.org/html/draft-ietf-avtext-client-to-mixer-audio-level-05
560 // with the intent that it can provide the RTP audio level indication.
561 //
562 // Frames passed to ProcessStream() with an |_energy| of zero are considered
563 // to have been muted. The RMS of the frame will be interpreted as -127.
564 virtual int RMS() = 0;
niklase@google.com470e71d2011-07-07 08:21:25 +0000565
566 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000567 virtual ~LevelEstimator() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000568};
569
570// The noise suppression (NS) component attempts to remove noise while
571// retaining speech. Recommended to be enabled on the client-side.
572//
573// Recommended to be enabled on the client-side.
574class NoiseSuppression {
575 public:
576 virtual int Enable(bool enable) = 0;
577 virtual bool is_enabled() const = 0;
578
579 // Determines the aggressiveness of the suppression. Increasing the level
580 // will reduce the noise level at the expense of a higher speech distortion.
581 enum Level {
582 kLow,
583 kModerate,
584 kHigh,
585 kVeryHigh
586 };
587
588 virtual int set_level(Level level) = 0;
589 virtual Level level() const = 0;
590
bjornv@webrtc.org08329f42012-07-12 21:00:43 +0000591 // Returns the internally computed prior speech probability of current frame
592 // averaged over output channels. This is not supported in fixed point, for
593 // which |kUnsupportedFunctionError| is returned.
594 virtual float speech_probability() const = 0;
595
niklase@google.com470e71d2011-07-07 08:21:25 +0000596 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000597 virtual ~NoiseSuppression() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000598};
599
600// The voice activity detection (VAD) component analyzes the stream to
601// determine if voice is present. A facility is also provided to pass in an
602// external VAD decision.
andrew@webrtc.orged083d42011-09-19 15:28:51 +0000603//
604// In addition to |stream_has_voice()| the VAD decision is provided through the
andrew@webrtc.org63a50982012-05-02 23:56:37 +0000605// |AudioFrame| passed to |ProcessStream()|. The |vad_activity_| member will be
andrew@webrtc.orged083d42011-09-19 15:28:51 +0000606// modified to reflect the current decision.
niklase@google.com470e71d2011-07-07 08:21:25 +0000607class VoiceDetection {
608 public:
609 virtual int Enable(bool enable) = 0;
610 virtual bool is_enabled() const = 0;
611
612 // Returns true if voice is detected in the current frame. Should be called
613 // after |ProcessStream()|.
614 virtual bool stream_has_voice() const = 0;
615
616 // Some of the APM functionality requires a VAD decision. In the case that
617 // a decision is externally available for the current frame, it can be passed
618 // in here, before |ProcessStream()| is called.
619 //
620 // VoiceDetection does _not_ need to be enabled to use this. If it happens to
621 // be enabled, detection will be skipped for any frame in which an external
622 // VAD decision is provided.
623 virtual int set_stream_has_voice(bool has_voice) = 0;
624
625 // Specifies the likelihood that a frame will be declared to contain voice.
626 // A higher value makes it more likely that speech will not be clipped, at
627 // the expense of more noise being detected as voice.
628 enum Likelihood {
629 kVeryLowLikelihood,
630 kLowLikelihood,
631 kModerateLikelihood,
632 kHighLikelihood
633 };
634
635 virtual int set_likelihood(Likelihood likelihood) = 0;
636 virtual Likelihood likelihood() const = 0;
637
638 // Sets the |size| of the frames in ms on which the VAD will operate. Larger
639 // frames will improve detection accuracy, but reduce the frequency of
640 // updates.
641 //
642 // This does not impact the size of frames passed to |ProcessStream()|.
643 virtual int set_frame_size_ms(int size) = 0;
644 virtual int frame_size_ms() const = 0;
645
646 protected:
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000647 virtual ~VoiceDetection() {}
niklase@google.com470e71d2011-07-07 08:21:25 +0000648};
649} // namespace webrtc
650
andrew@webrtc.orgd72b3d62012-11-15 21:46:06 +0000651#endif // WEBRTC_MODULES_AUDIO_PROCESSING_INCLUDE_AUDIO_PROCESSING_H_