Xin Zhang commited on
Commit ·
e03f21e
1
Parent(s): 418e2a0
Update model and processor files
Browse files
moyoyo_asr_models/ggml-small.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 487601984
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:951596a31b1c96a01b7a2b1bc511f665d900c679126134f6ec18db5ec4a485fe
|
| 3 |
size 487601984
|
transcribe/helpers/vadprocessor.py
CHANGED
|
@@ -137,7 +137,7 @@ class VADIteratorOnnx:
|
|
| 137 |
return_seconds: bool (default - False)
|
| 138 |
whether return timestamps in seconds (default - samples)
|
| 139 |
"""
|
| 140 |
-
|
| 141 |
window_size_samples = 512 if self.sampling_rate == 16000 else 256
|
| 142 |
x = x[:window_size_samples]
|
| 143 |
if len(x) < window_size_samples:
|
|
@@ -156,7 +156,7 @@ class VADIteratorOnnx:
|
|
| 156 |
speech_start = max(0, self.current_sample - window_size_samples)
|
| 157 |
self.start = speech_start
|
| 158 |
return {'start': int(speech_start) if not return_seconds else round(speech_start / self.sampling_rate, 1)}
|
| 159 |
-
|
| 160 |
if (speech_prob >= self.threshold) and self.current_sample - self.start >= self.max_speech_samples:
|
| 161 |
if self.temp_end:
|
| 162 |
self.temp_end = 0
|
|
@@ -175,7 +175,7 @@ class VADIteratorOnnx:
|
|
| 175 |
return {'end': int(speech_end) if not return_seconds else round(speech_end / self.sampling_rate, 1)}
|
| 176 |
|
| 177 |
return None
|
| 178 |
-
|
| 179 |
|
| 180 |
|
| 181 |
class VadV2:
|
|
@@ -267,15 +267,15 @@ class VadV2:
|
|
| 267 |
|
| 268 |
return result
|
| 269 |
return None
|
| 270 |
-
|
| 271 |
|
| 272 |
-
|
|
|
|
| 273 |
class VadProcessor:
|
| 274 |
def __init__(
|
| 275 |
self,
|
| 276 |
prob_threshold=0.5,
|
| 277 |
-
silence_s=0.
|
| 278 |
-
cache_s=0.
|
| 279 |
sr=16000
|
| 280 |
):
|
| 281 |
self.prob_thres = prob_threshold
|
|
@@ -284,7 +284,7 @@ class VadProcessor:
|
|
| 284 |
self.silence_s = silence_s
|
| 285 |
|
| 286 |
self.vad = VadV2(self.prob_thres, self.sr, self.silence_s * 1000, self.cache_s * 1000, max_speech_duration_s=15)
|
| 287 |
-
|
| 288 |
|
| 289 |
def process_audio(self, audio_buffer: np.ndarray):
|
| 290 |
audio = np.array([], np.float32)
|
|
|
|
| 137 |
return_seconds: bool (default - False)
|
| 138 |
whether return timestamps in seconds (default - samples)
|
| 139 |
"""
|
| 140 |
+
|
| 141 |
window_size_samples = 512 if self.sampling_rate == 16000 else 256
|
| 142 |
x = x[:window_size_samples]
|
| 143 |
if len(x) < window_size_samples:
|
|
|
|
| 156 |
speech_start = max(0, self.current_sample - window_size_samples)
|
| 157 |
self.start = speech_start
|
| 158 |
return {'start': int(speech_start) if not return_seconds else round(speech_start / self.sampling_rate, 1)}
|
| 159 |
+
|
| 160 |
if (speech_prob >= self.threshold) and self.current_sample - self.start >= self.max_speech_samples:
|
| 161 |
if self.temp_end:
|
| 162 |
self.temp_end = 0
|
|
|
|
| 175 |
return {'end': int(speech_end) if not return_seconds else round(speech_end / self.sampling_rate, 1)}
|
| 176 |
|
| 177 |
return None
|
| 178 |
+
|
| 179 |
|
| 180 |
|
| 181 |
class VadV2:
|
|
|
|
| 267 |
|
| 268 |
return result
|
| 269 |
return None
|
|
|
|
| 270 |
|
| 271 |
+
|
| 272 |
+
|
| 273 |
class VadProcessor:
|
| 274 |
def __init__(
|
| 275 |
self,
|
| 276 |
prob_threshold=0.5,
|
| 277 |
+
silence_s=0.2,
|
| 278 |
+
cache_s=0.15,
|
| 279 |
sr=16000
|
| 280 |
):
|
| 281 |
self.prob_thres = prob_threshold
|
|
|
|
| 284 |
self.silence_s = silence_s
|
| 285 |
|
| 286 |
self.vad = VadV2(self.prob_thres, self.sr, self.silence_s * 1000, self.cache_s * 1000, max_speech_duration_s=15)
|
| 287 |
+
|
| 288 |
|
| 289 |
def process_audio(self, audio_buffer: np.ndarray):
|
| 290 |
audio = np.array([], np.float32)
|