public:
TRANSLATOR_DECLARATIONS (Dynamic_performer);
protected:
+ virtual void acknowledge_audio_element (Audio_element_info info);
virtual void finalize ();
void stop_translation_timestep ();
void process_music ();
void listen_absolute_dynamic (Stream_event *);
private:
+ void close_and_enqueue_span ();
+ Real compute_departure_volume (Direction depart_dir,
+ Real start_vol,
+ Real end_vol,
+ Real min_vol,
+ Real max_vol);
+ bool drive_state_machine (Direction next_grow_dir);
// next_vol < 0 means select a target dynamic based on growth direction.
// return actual next volume (computed if not provided)
- Real end_span (Real next_vol = -1.0);
+ Real finish_queued_spans (Real next_vol = -1.0);
+ Real look_up_absolute_volume (SCM dynamicString,
+ Real defaultValue);
private:
+ // This performer queues a number of dynamic spans waiting for the following
+ // pattern before computing their volume levels.
+ //
+ // 1. the first (de)crescendo, followed by ...
+ // 2. zero or more spans that either change in the same direction as the
+ // first or do not change, followed by ...
+ // 3. zero or more spans that either change in the opposite direction as the
+ // first or do not change
+ //
+ // The search may be cut short by an absolute dynamic or the end of the
+ // context.
+ enum State
+ {
+ STATE_INITIAL = 0, // waiting for a (de)crescendo
+ STATE_DEPART, // enqueued the first span, gathering same-direction spans
+ STATE_RETURN // gathering opposite-direction spans
+ };
+
+ struct UnfinishedSpan
+ {
+ Audio_span_dynamic *dynamic_;
+ Direction grow_dir_;
+
+ UnfinishedSpan () : dynamic_ (0), grow_dir_ (CENTER) {}
+ };
+
+ struct DynamicQueue
+ {
+ vector<UnfinishedSpan> spans_;
+ // total duration of (de)crescendi (i.e. excluding fixed-volume spans)
+ Real change_duration_;
+ Real min_target_vol_;
+ Real max_target_vol_;
+
+ DynamicQueue () : change_duration_ (0) {}
+
+ void clear ()
+ {
+ spans_.clear ();
+ change_duration_ = 0;
+ }
+
+ void push_back (const UnfinishedSpan &span,
+ Real min_target_vol,
+ Real max_target_vol)
+ {
+ if (span.grow_dir_ != CENTER)
+ change_duration_ += span.dynamic_->get_duration ();
+ min_target_vol_ = min_target_vol;
+ max_target_vol_ = max_target_vol;
+ spans_.push_back (span);
+ }
+
+ void set_volume (Real start_vol, Real target_vol);
+ };
+
+private:
+ vector<Audio_note *> notes_;
Stream_event *script_event_;
Drul_array<Stream_event *> span_events_;
Direction next_grow_dir_;
- Audio_span_dynamic *span_dynamic_;
- Direction grow_dir_; // of span_dynamic_
+ Direction depart_dir_;
+ UnfinishedSpan open_span_;
+ DynamicQueue depart_queue_;
+ DynamicQueue return_queue_;
+ State state_;
};
-Dynamic_performer::Dynamic_performer ()
+Dynamic_performer::Dynamic_performer (Context *c)
+ : Performer (c),
+ script_event_ (0),
+ next_grow_dir_ (CENTER),
+ depart_dir_ (CENTER),
+ state_ (STATE_INITIAL)
{
- script_event_ = 0;
span_events_[LEFT]
- = span_events_[RIGHT] = 0;
- next_grow_dir_ = CENTER;
- span_dynamic_ = 0;
- grow_dir_ = CENTER;
+ = span_events_[RIGHT] = 0;
+}
+
+void
+Dynamic_performer::acknowledge_audio_element (Audio_element_info inf)
+{
+ // Keep track of the notes played in this translation time step so that they
+ // can be pointed to the current dynamic in stop_translation_timestep.
+ if (Audio_note *n = dynamic_cast<Audio_note *> (inf.elem_)) {
+ notes_.push_back (n);
+ }
+}
+
+bool
+Dynamic_performer::drive_state_machine (Direction next_grow_dir)
+{
+ switch (state_)
+ {
+ case STATE_INITIAL:
+ if (next_grow_dir != CENTER)
+ {
+ state_ = STATE_DEPART;
+ depart_dir_ = next_grow_dir;
+ }
+ break;
+
+ case STATE_DEPART:
+ if (next_grow_dir == -depart_dir_)
+ state_ = STATE_RETURN;
+ break;
+
+ case STATE_RETURN:
+ if (next_grow_dir == depart_dir_)
+ {
+ state_ = STATE_DEPART;
+ return true;
+ }
+ break;
+ }
+
+ return false;
+}
+
+void
+Dynamic_performer::close_and_enqueue_span ()
+{
+ if (!open_span_.dynamic_)
+ programming_error ("no open dynamic span");
+ else
+ {
+ DynamicQueue &dq
+ = (state_ == STATE_RETURN) ? return_queue_ : depart_queue_;
+
+ // Changing equalizer settings in the course of the performance does not
+ // seem very likely. This is a fig leaf: Equalize these limit volumes
+ // now as the required context properties are current. Note that only
+ // the limits at the end of the last span in the queue are kept.
+
+ // Resist diminishing to silence. (Idea: Look up "ppppp"
+ // with dynamicAbsoluteVolumeFunction, however that would yield 0.25.)
+ const Real min_target = equalize_volume (0.1);
+ const Real max_target
+ = equalize_volume (Audio_span_dynamic::MAXIMUM_VOLUME);
+
+ open_span_.dynamic_->set_end_moment (now_mom ());
+ dq.push_back (open_span_, min_target, max_target);
+ }
+
+ open_span_ = UnfinishedSpan ();
+}
+
+// Set the starting and target volume for each span in the queue. The gain
+// (loss) of any (de)crescendo is proportional to its share of the total time
+// spent changing.
+void
+Dynamic_performer::DynamicQueue::set_volume (Real start_vol,
+ Real target_vol)
+{
+ const Real gain = target_vol - start_vol;
+ Real dur = 0; // duration of (de)crescendi processed so far
+ Real vol = start_vol;
+ for (vector<UnfinishedSpan>::iterator it = spans_.begin ();
+ it != spans_.end (); ++it)
+ {
+ const Real prev_vol = vol;
+ if (it->grow_dir_ != CENTER)
+ {
+ // grant this (de)crescendo its portion of the gain
+ dur += it->dynamic_->get_duration ();
+ vol = start_vol + gain * (dur / change_duration_);
+ }
+ it->dynamic_->set_volume (prev_vol, vol);
+ }
+}
+
+// Return a volume which is reasonably distant from the given start and end
+// volumes in the given direction, for use as a peak volume in a passage with a
+// crescendo followed by a decrescendo (or vice versa). If the given volumes
+// are equal, the returned volume is a also reasonable target volume for a
+// single (de)crescendo.
+//
+// The given minimum and maximum volumes are the allowable dynamic range.
+Real
+Dynamic_performer::compute_departure_volume (Direction depart_dir,
+ Real start_vol,
+ Real end_vol,
+ Real min_vol,
+ Real max_vol)
+{
+ if (depart_dir == CENTER)
+ return start_vol;
+
+ // Try to find a volume that is a minimum distance from the starting and
+ // ending volumes. If the endpoint volumes differ, the nearer one is padded
+ // less than the farther one.
+ //
+ // Example: mf < ... > p. The legacy behavior was to use a 25% of the
+ // dynamic range for a (de)crescendo to an unspecified target, and this tries
+ // to preserve that, but is not possible to use a 25% change for both the
+ // crescendo and the decrescendo and meet the constraints of this example.
+ // The decrescendo is a greater change than the crescendo. Believing that
+ // 25% is already more than enough for either, pad using 25% for the greater
+ // change and 7% for the lesser change.
+ //
+ // Idea: Use a context property or callback, e.g. the difference between two
+ // dynamics in dynamicAbsoluteVolumeFunction. 0.25 is the default difference
+ // between "p" and "ff". (Isn't that rather wide for this purpose?) 0.07 is
+ // the default difference between "mp" and "mf".
+ const Real far_padding = 0.25;
+ const Real near_padding = 0.07;
+
+ // If for some reason one of the endpoints is already below the supposed
+ // minimum or maximum, just accept it.
+ min_vol = min (min (min_vol, start_vol), end_vol);
+ max_vol = max (max (max_vol, start_vol), end_vol);
+
+ const Real vol_range = max_vol - min_vol;
+
+ const Real near_vol = minmax (depart_dir, start_vol, end_vol)
+ + depart_dir * near_padding * vol_range;
+ const Real far_vol = minmax (-depart_dir, start_vol, end_vol)
+ + depart_dir * far_padding * vol_range;
+ const Real depart_vol = minmax (depart_dir, near_vol, far_vol);
+ return max (min (depart_vol, max_vol), min_vol);
}
-Real Dynamic_performer::end_span (Real next_vol)
+Real
+Dynamic_performer::finish_queued_spans (Real next_vol)
{
- if (!span_dynamic_)
+ if (depart_queue_.spans_.empty ())
{
- programming_error("no dynamic span to end");
+ programming_error ("no dynamic span to finish");
return next_vol;
}
- Real start_vol = span_dynamic_->get_start_volume ();
- Real target_vol = start_vol;
-
- if (grow_dir_ != CENTER) {
- // If the target dynamic is not specified, grow to a reasonable target
- // in the desired direction. Do the same for cases like mf < p.
- //
- // TODO To improve on this, keep a queue of Audio_span_dynamics and compute
- // multiple intermediate targets based on the next explicit dynamic.
- // Consider cases like mf < ... < ff with only mf and ff specified.
- // Consider growing in proportion to the duration of each (de)crescendo in
- // the sequence, which may be separated by spans with no change in volume.
- if ((next_vol < 0) || (sign(next_vol - start_vol) != grow_dir_))
- {
- Real min_vol = equalize_volume (0.1);
- Real max_vol = equalize_volume (Audio_span_dynamic::MAXIMUM_VOLUME);
- target_vol = max (min (start_vol + grow_dir_ * (max_vol - min_vol) * 0.25, max_vol), min_vol);
- }
- else
- {
- target_vol = next_vol;
- }
- }
+ const Real start_vol = depart_queue_.spans_.front ().dynamic_->get_start_volume ();
+
+ if (return_queue_.spans_.empty ())
+ {
+ Real depart_vol = next_vol;
- span_dynamic_->set_end_moment (now_mom ());
- span_dynamic_->set_volume (start_vol, target_vol);
- span_dynamic_ = 0;
+ // If the next dynamic is not specified or is inconsistent with the
+ // direction of growth, choose a reasonable target.
+ if ((next_vol < 0) || (depart_dir_ != sign (next_vol - start_vol)))
+ {
+ depart_vol = compute_departure_volume (depart_dir_,
+ start_vol, start_vol,
+ depart_queue_.min_target_vol_,
+ depart_queue_.max_target_vol_);
+ }
- return (next_vol >= 0) ? next_vol : target_vol;
+ depart_queue_.set_volume (start_vol, depart_vol);
+ depart_queue_.clear ();
+ return (next_vol >= 0) ? next_vol : depart_vol;
+ }
+ else
+ {
+ // If the next dynamic is not specified, return to the starting volume.
+ const Real return_vol = (next_vol >= 0) ? next_vol : start_vol;
+ Real depart_vol = compute_departure_volume (depart_dir_,
+ start_vol, return_vol,
+ depart_queue_.min_target_vol_,
+ depart_queue_.max_target_vol_);
+ depart_queue_.set_volume (start_vol, depart_vol);
+ depart_queue_.clear ();
+ return_queue_.set_volume (depart_vol, return_vol);
+ return_queue_.clear ();
+ return return_vol;
+ }
}
Real
void
Dynamic_performer::finalize ()
{
- if (span_dynamic_)
- {
- end_span ();
- }
+ if (open_span_.dynamic_)
+ close_and_enqueue_span ();
+ finish_queued_spans ();
+}
+
+Real
+Dynamic_performer::look_up_absolute_volume (SCM dynamicString,
+ Real defaultValue)
+{
+ SCM proc = get_property ("dynamicAbsoluteVolumeFunction");
+
+ SCM svolume = SCM_EOL;
+ if (ly_is_procedure (proc))
+ svolume = scm_call_1 (proc, dynamicString);
+
+ return robust_scm2double (svolume, defaultValue);
}
void
{
Real volume = -1;
- if (script_event_)
+ if (script_event_) // explicit dynamic
{
- // Explicit dynamic script event: determine the volume.
- SCM proc = get_property ("dynamicAbsoluteVolumeFunction");
-
- SCM svolume = SCM_EOL;
- if (ly_is_procedure (proc))
- {
- // urg
- svolume = scm_call_1 (proc, script_event_->get_property ("text"));
- }
-
- volume = equalize_volume (robust_scm2double (svolume, Audio_span_dynamic::DEFAULT_VOLUME));
+ volume = look_up_absolute_volume (script_event_->get_property ("text"),
+ Audio_span_dynamic::DEFAULT_VOLUME);
+ volume = equalize_volume (volume);
}
- else if (!span_dynamic_) // first time through
+ else if (!open_span_.dynamic_) // first time only
{
+ // Idea: look_up_absolute_volume (ly_symbol2scm ("mf")).
+ // It is likely to change regtests.
volume = equalize_volume (Audio_span_dynamic::DEFAULT_VOLUME);
}
// end the current span at relevant points
- if (span_dynamic_
+ if (open_span_.dynamic_
&& (span_events_[START] || span_events_[STOP] || script_event_))
{
- volume = end_span (volume);
+ close_and_enqueue_span ();
+ if (script_event_)
+ {
+ state_ = STATE_INITIAL;
+ volume = finish_queued_spans (volume);
+ }
}
// start a new span so that some dynamic is always in effect
- if (!span_dynamic_)
+ if (!open_span_.dynamic_)
{
- Stream_event *cause =
- span_events_[START] ? span_events_[START] :
- script_event_ ? script_event_ :
- span_events_[STOP];
-
- span_dynamic_ = new Audio_span_dynamic (now_mom (), volume);
- grow_dir_ = next_grow_dir_;
- announce_element (Audio_element_info (span_dynamic_, cause));
+ if (drive_state_machine (next_grow_dir_))
+ volume = finish_queued_spans (volume);
+
+ // if not known by now, use a default volume for robustness
+ if (volume < 0)
+ volume = equalize_volume (Audio_span_dynamic::DEFAULT_VOLUME);
+
+ Stream_event *cause
+ = span_events_[START] ? span_events_[START]
+ : script_event_ ? script_event_
+ : span_events_[STOP];
+
+ open_span_.dynamic_ = new Audio_span_dynamic (now_mom (), volume);
+ open_span_.grow_dir_ = next_grow_dir_;
+ announce_element (Audio_element_info (open_span_.dynamic_, cause));
}
}
void
Dynamic_performer::stop_translation_timestep ()
{
+ // link notes to the current dynamic
+ if (!open_span_.dynamic_)
+ programming_error("no current dynamic");
+ else
+ {
+ for (vector<Audio_note *>::const_iterator ni = notes_.begin ();
+ ni != notes_.end (); ++ni)
+ {
+ (*ni)->dynamic_ = open_span_.dynamic_;
+ }
+ }
+ notes_.clear ();
+
script_event_ = 0;
span_events_[LEFT]
- = span_events_[RIGHT] = 0;
+ = span_events_[RIGHT] = 0;
next_grow_dir_ = CENTER;
}