- use native speech detector

src/agent_config.py

@@ -36,8 +36,8 @@ class AgentConfig:
     # Should the first task run immediately ?
     ForceRun = False
 
-    # Use silence eraser or not (speech detector is used in this case)
+    # Use external speech detector if needed
-    UseSilenceEraser = True
+    UseSpeechDetector = False
 
     # Path to log file
     LogPath : Path = None
@@ -90,7 +90,7 @@ class AgentConfig:
 
             if 'speech_detector' in config:
                 if config['speech_detector']:
-                    self.UseSilenceEraser = False
+                    self.UseSpeechDetector = True
 
             if 'audio' in config:
                 audio = config['audio']

src/agent_gsm.py

@@ -68,7 +68,10 @@ def detect_degraded_signal(file_test: Path, file_reference: Path) -> SignalBound
         # Seems some problem with recording, return zero boundaries
         return SignalBoundaries()
     
-    r = bt_signal.find_reference_signal(file_test)
+    if CONFIG.UseSpeechDetector:
+        r = bt_signal.find_reference_signal_via_speechdetector(file_test)
+    else:
+        r = bt_signal.find_reference_signal(file_test)
     
     if r.offset_start == 0.0 and is_caller:
         r.offset_start = 5.0 # Skip ringing tones
@@ -78,7 +81,10 @@ def detect_degraded_signal(file_test: Path, file_reference: Path) -> SignalBound
 
 def detect_reference_signal(file_reference: Path) -> SignalBoundaries:
     # Run silence eraser on reference file as well
-    result = bt_signal.find_reference_signal(file_reference)
+    if CONFIG.UseSpeechDetector:
+        result = bt_signal.find_reference_signal_via_speechdetector(file_reference)
+    else:
+        result = bt_signal.find_reference_signal(file_reference)
     return result
 
 

src/bt_signal.py

@@ -3,33 +3,22 @@
 import sys
 import os
 import pathlib
+from utils_types import SignalBoundaries
+from utils_sevana import speech_detector
 
 from pydub import silence, AudioSegment
 
-class SignalBoundaries:
+SILENCE_DELTA = 16
-    # Offset from start (in seconds)
-    offset_start: float
-
-    # Offset from finish (in seconds)
-    offset_finish: float
-
-    def __init__(self, offset_start = 0.0, offset_finish = 0.0) -> None:
-        self.offset_start = offset_start
-        self.offset_finish = offset_finish
-
-    def __repr__(self) -> str:
-        return f'[offset_start: {round(self.offset_start, 3)}, offset_finish : {round(self.offset_finish, 3)}]'
-
 
 def find_reference_signal(input_file: pathlib.Path, output_file: pathlib.Path = None, use_end_offset: bool = True) -> SignalBoundaries:
     myaudio = AudioSegment.from_wav(str(input_file))
     dBFS = myaudio.dBFS
     
     # Find silence intervals
-    intervals = silence.detect_nonsilent(myaudio, min_silence_len=1000, silence_thresh=dBFS-17, seek_step=50)
+    intervals = silence.detect_nonsilent(myaudio, min_silence_len=1000, silence_thresh=dBFS-SILENCE_DELTA, seek_step=50)
     
     # Translate to seconds
-    intervals = [((start/1000),(stop/1000)) for start,stop in intervals] #in sec
+    intervals = [((start/1000),(stop/1000)) for start,stop in intervals] # in sec
 
     # print(intervals)
 
@@ -48,6 +37,12 @@ def find_reference_signal(input_file: pathlib.Path, output_file: pathlib.Path =
 
     return SignalBoundaries()
 
+
+def find_reference_signal_via_speechdetector(input_file: pathlib.Path) -> SignalBoundaries:
+    bounds = speech_detector(str(input_file))
+    r = SignalBoundaries(bounds[0], bounds[1])
+    return bounds
+
 if __name__ == '__main__':
     if len(sys.argv) < 2:
         print(f'Please specify input filename.')

src/utils_types.py

@@ -6,12 +6,25 @@ import utils
 import json
 from crontab import CronTab
 
-
-
 # Exit codes
 EXIT_OK = 0
 EXIT_ERROR = 1
 
+class SignalBoundaries:
+    # Offset from start (in seconds)
+    offset_start: float
+
+    # Offset from finish (in seconds)
+    offset_finish: float
+
+    def __init__(self, offset_start = 0.0, offset_finish = 0.0) -> None:
+        self.offset_start = offset_start
+        self.offset_finish = offset_finish
+
+    def __repr__(self) -> str:
+        return f'[offset_start: {round(self.offset_start, 3)}, offset_finish : {round(self.offset_finish, 3)}]'
+
+
 class Phone:
     identifier: int = 0
     name: str = ""