enh: offset voice key, version 0.5

psychopy/voicekey/__init__.py

@@ -5,18 +5,16 @@
 
 Copyright (c) Jeremy R. Gray, 2015
 License: Distributed under the terms of the GPLv3
-Version: 0.4
 Dev status: beta. Can work well in some circumstances, not widely tested.
-Smoother with 64-bit python and pyo.
 
 _BaseVoiceKey is the main abstract class. Subclass and override the detect()
-method. See SimpleThresholdVoiceKey
-
-See readme.txt for notes, vk_plot.py for demos and testing.
+method. See SimpleThresholdVoiceKey or OnsetVoiceKey for examples.
 """
 
 from __future__ import division
 
+__version__ = 0.5
+
 import sys
 import os
 import numpy as np
@@ -32,8 +30,7 @@
 # pyo_server will point to a booted pyo server once pyo_init() is called:
 pyo_server = None
 
-# Various bits and pieces:
-from . signal import _BaseVoiceKeySignal
+# helper functions for time, signal processing, and file I/O:
 from . vk_tools import *
 
 # Constants:
@@ -207,7 +204,7 @@ def _set_signaler(self):
     def _set_tables(self):
         """Set up the pyo tables (allocate memory, etc).
 
-        One source -> three pyo tables: chunk=short, whole=all, baseline
+        One source -> three pyo tables: chunk=short, whole=all, baseline.
         triggers fill tables from self._source; make triggers in .start()
         """
         sec_per_chunk = self.msPerChunk / 1000.
@@ -439,9 +436,10 @@ def wait_for_event(self, plus=0):
             if naptime > 0:
                 sleep(naptime)
             self.stop()
-        # next sleep() helps avoid pyo "ReferenceError: weakly-referenced
-        #                               object no longer exists"
+        # next sleep() helps avoid pyo error:
+        #    "ReferenceError: weakly-referenced object no longer exists"
         sleep(1.5 * self.msPerChunk / 1000.)
+
         return self.elapsed
 
     def save(self, ftype='', dtype='int16'):
@@ -515,62 +513,46 @@ def detect(self):
             self.trip()
 
 
-class VoicelessPlosiveVoiceKey(_BaseVoiceKey):
-    """Class to detect and signal the offset of a vowel followed by a voiceless
-    plosive, e.g. the end of "ah" in utterance "ah pa".
+class OffsetVoiceKey(_BaseVoiceKey):
+    """Class to detect the offset of a single-word utterance.
+
+    Ends the recording after a delay; default = 300ms later.
     """
-    def __init__(self, sec=1.5, msPerChunk=2, file_out='', file_in='',
-                 duration=0.070, proportion=0.7, signaler=None,
-                 start=0, stop=-1, baseline=0):
+    def __init__(self, sec=10, file_out='', file_in='', delay=0.3, **kwargs):
         """Adjust parameters `duration` and `proportion` as needed.
         """
         config = {'sec': sec,
-                  'msPerChunk': msPerChunk,
                   'file_out': file_out,
                   'file_in': file_in,
-                  'duration': duration,  # min duration of vowel, in sec
-                  'proportion': proportion,  # min prop of chunks > threshold
-                  'signaler': signaler,  # obj for obj.signal() upon event
-                  'start': start,
-                  'stop': stop,
-                  'baseline': baseline,
+                  'delay': delay,
                  }
-        super(VoicelessPlosiveVoiceKey, self).__init__(**config)
+        kwargs.update(config)
+        super(OffsetVoiceKey, self).__init__(**kwargs)
 
     def detect(self):
-        """Detect the near-end of the first sustained speech-like sound.
-
-        Called every chunk, so keep it efficient.
-
-        Define multiple conditions. Trip (= trigger the event) if all are met.
-        - minimum time has elapsed (baseline period)
-        - have gone above a minimum threshold recently
-        - met that threshold for some proportion of recent chunks (hold time)
-        - sound is currently greatly attenuated (trailing edge)
+        """Wait for onset, offset, delay, then end the recording.
         """
-
         if self.event_detected or not self.baseline:
             return
-
-        thr_norm = 0.03 * self.max_bp  # 3% of recent max value; not ensured to be recent
-        if not hasattr(self, '_hold'):
-            # compute once, cache
-            self._hold = -1 * int(self.config['duration'] * 1000. / self.msPerChunk)
-            self._offset = -1 * int(8. / self.msPerChunk)  # ms -> chunks
-
-        vals = self.power_bp[self._hold:self._offset]
-        max_val = np.max(vals)
-        prop_over_thr = np.mean(vals > thr_norm)  # mean of 0, 1's
-        loud_enough = max_val > 5 * self.baseline and max_val > 500
-
-        recent = np.mean(self.power_bp[self._offset:])
-        quiet_recently = recent < 2 * thr_norm
-
-        conditions = (prop_over_thr > self.config['proportion'],
-                      loud_enough,
-                      quiet_recently)
-        if all(conditions):
+        if not self.event_onset:
+            window = 5  # chunks
+            threshold = 10 * self.baseline
+            conditions = all([x > threshold for x in self.power_bp[-window:]])
+            if conditions:
+                self.event_lag = window * self.msPerChunk / 1000.
+                self.event_onset = self.elapsed - self.event_lag
+                self.event_offset = 0
+        elif not self.event_offset:
+            window = 25
+            threshold = 10 * self.baseline
+            conditions = all([x < threshold for x in self.power_bp[-window:]])
+            if conditions:
+                self.event_lag = window * self.msPerChunk / 1000.
+                self.event_offset = self.elapsed - self.event_lag
+                self.event_time = self.event_offset  # for plotting
+        elif self.elapsed > self.event_offset + self.config['delay']:
             self.trip()
+            self.stop()
 
 
 ### ----- Convenience classes -------------------------------------------------