Commit @175c382b7b1663fdee6316f039426d4c93c27e44 - yjyoon/whisper_streaming

Dominik Macháček 2024-02-07

increasing timestamps fixed

but the code needs to be simplified and cleaned before merging

@175c382b7b1663fdee6316f039426d4c93c27e44

1c523f2

175c382

whisper_online.py

--- whisper_online.py

+++ whisper_online.py


 
         self.buffer_trimming_way, self.buffer_trimming_sec = buffer_trimming
 
-    def init(self):
+    def init(self, keep_offset=False):
         """run this when starting or restarting processing"""
         self.audio_buffer = np.array([],dtype=np.float32)
-        self.buffer_time_offset = 0
-
         self.transcript_buffer = HypothesisBuffer(logfile=self.logfile)
+        if not keep_offset:
+            self.buffer_time_offset = 0
+            self.transcript_buffer.last_commited_time = 0
+        else:
+            self.transcript_buffer.last_commited_time = self.buffer_time_offset
+
         self.commited = []
         self.last_chunked_at = 0
 
-        self.silence_iters = 0
 
     def insert_audio_chunk(self, audio):
         self.audio_buffer = np.append(self.audio_buffer, audio)

         o = self.transcript_buffer.complete()
         f = self.to_flush(o)
         print("last, noncommited:",f,file=self.logfile)
+        self.buffer_time_offset += len(self.audio_buffer)/16000
         return f
 
 

1c523f2

175c382

whisper_online_vac.py

--- whisper_online_vac.py

+++ whisper_online_vac.py


 
     def finish(self):
         ret = self.online.finish()
-        self.online.init()
+        self.online.init(keep_offset=True)
         self.current_online_chunk_buffer_size = 0
         return ret
 

Add a comment

Open 0
Closed 0

List

...	...	@@ -235,16 +235,19 @@
235	235
236	236	self.buffer_trimming_way, self.buffer_trimming_sec = buffer_trimming
237	237
238		- def init(self):
	238	+ def init(self, keep_offset=False):
239	239	"""run this when starting or restarting processing"""
240	240	self.audio_buffer = np.array([],dtype=np.float32)
241		- self.buffer_time_offset = 0
242		-
243	241	self.transcript_buffer = HypothesisBuffer(logfile=self.logfile)
	242	+ if not keep_offset:
	243	+ self.buffer_time_offset = 0
	244	+ self.transcript_buffer.last_commited_time = 0
	245	+ else:
	246	+ self.transcript_buffer.last_commited_time = self.buffer_time_offset
	247	+
244	248	self.commited = []
245	249	self.last_chunked_at = 0
246	250
247		- self.silence_iters = 0
248	251
249	252	def insert_audio_chunk(self, audio):
250	253	self.audio_buffer = np.append(self.audio_buffer, audio)
...	...	@@ -400,6 +403,7 @@
400	403	o = self.transcript_buffer.complete()
401	404	f = self.to_flush(o)
402	405	print("last, noncommited:",f,file=self.logfile)
	406	+ self.buffer_time_offset += len(self.audio_buffer)/16000
403	407	return f
404	408
405	409

...	...	@@ -45,7 +45,7 @@
45	45
46	46	def finish(self):
47	47	ret = self.online.finish()
48		- self.online.init()
	48	+ self.online.init(keep_offset=True)
49	49	self.current_online_chunk_buffer_size = 0
50	50	return ret
51	51

Delete comment