Update model and processor files

moyoyo_asr_models/ggml-small.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f6ef171491de375b741059400ba9a0aead023122b7a7db731b4943f9baa0f97
+oid sha256:951596a31b1c96a01b7a2b1bc511f665d900c679126134f6ec18db5ec4a485fe
 size 487601984

transcribe/helpers/vadprocessor.py

@@ -137,7 +137,7 @@ class VADIteratorOnnx:
         return_seconds: bool (default - False)
             whether return timestamps in seconds (default - samples)
         """
-        
+
         window_size_samples = 512 if self.sampling_rate == 16000 else 256
         x = x[:window_size_samples]
         if len(x) < window_size_samples:
@@ -156,7 +156,7 @@ class VADIteratorOnnx:
             speech_start = max(0, self.current_sample - window_size_samples)
             self.start = speech_start
             return {'start': int(speech_start) if not return_seconds else round(speech_start / self.sampling_rate, 1)}
-        
+
         if (speech_prob >= self.threshold) and self.current_sample - self.start >= self.max_speech_samples:
             if self.temp_end:
                 self.temp_end = 0
@@ -175,7 +175,7 @@ class VADIteratorOnnx:
                 return {'end': int(speech_end) if not return_seconds else round(speech_end / self.sampling_rate, 1)}
 
         return None
-    
+
 
 
 class VadV2:
@@ -267,15 +267,15 @@ class VadV2:
 
             return result
         return None
-    
 
-    
+
+
 class VadProcessor:
     def __init__(
             self,
             prob_threshold=0.5,
-            silence_s=0.3,
-            cache_s=0.25,
+            silence_s=0.2,
+            cache_s=0.15,
             sr=16000
     ):
         self.prob_thres = prob_threshold
@@ -284,7 +284,7 @@ class VadProcessor:
         self.silence_s = silence_s
 
         self.vad = VadV2(self.prob_thres, self.sr, self.silence_s * 1000, self.cache_s * 1000, max_speech_duration_s=15)
-    
+
 
     def process_audio(self, audio_buffer: np.ndarray):
         audio = np.array([], np.float32)