Spaces:
Building
Building
Update stt/stt_deepgram.py
Browse files- stt/stt_deepgram.py +25 -25
stt/stt_deepgram.py
CHANGED
@@ -262,49 +262,49 @@ class DeepgramSTT(STTInterface):
|
|
262 |
"""Stream audio chunk and get transcription results"""
|
263 |
if not self.is_streaming or not self.live_connection:
|
264 |
raise RuntimeError("Streaming not started. Call start_streaming() first.")
|
265 |
-
|
266 |
-
# Don't send audio if final result already received
|
267 |
-
if self.final_result_received:
|
268 |
-
log_debug("Final result already received, ignoring audio chunk")
|
269 |
-
return
|
270 |
-
|
271 |
try:
|
272 |
-
#
|
273 |
if self.total_chunks < 3:
|
274 |
-
# Linear16 formatı kontrolü
|
275 |
if len(audio_chunk) >= 4:
|
276 |
-
# İlk 2 byte'ı int16 olarak oku
|
277 |
import struct
|
278 |
try:
|
279 |
-
# Linear16 ise ilk sample'ı okuyabilmeliyiz
|
280 |
first_sample = struct.unpack('<h', audio_chunk[:2])[0]
|
281 |
log_info(f"🔊 Audio format check - Chunk #{self.total_chunks}: First sample={first_sample}, Size={len(audio_chunk)} bytes")
|
282 |
-
|
283 |
-
# Eğer WebM/Opus ise magic bytes farklı olur
|
284 |
-
if audio_chunk[:4] == b'webm' or audio_chunk[:4] == b'\x1a\x45\xdf\xa3':
|
285 |
-
log_error("❌ WebM format detected instead of Linear16!")
|
286 |
except:
|
287 |
log_warning("⚠️ Could not parse as Linear16")
|
288 |
-
|
289 |
-
# Send audio to Deepgram
|
290 |
-
self.live_connection.send(audio_chunk)
|
291 |
-
|
292 |
-
self.total_chunks += 1
|
293 |
-
self.total_audio_bytes += len(audio_chunk)
|
294 |
|
295 |
-
#
|
296 |
-
if self.
|
297 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
298 |
|
299 |
-
#
|
300 |
while True:
|
301 |
try:
|
302 |
result = self.responses_queue.get_nowait()
|
|
|
|
|
|
|
|
|
303 |
if result.is_final:
|
|
|
|
|
|
|
|
|
|
|
304 |
yield result
|
|
|
305 |
except queue.Empty:
|
306 |
break
|
307 |
-
|
308 |
except Exception as e:
|
309 |
log_error(f"❌ Error streaming audio", error=str(e))
|
310 |
self.is_streaming = False
|
|
|
262 |
"""Stream audio chunk and get transcription results"""
|
263 |
if not self.is_streaming or not self.live_connection:
|
264 |
raise RuntimeError("Streaming not started. Call start_streaming() first.")
|
265 |
+
|
|
|
|
|
|
|
|
|
|
|
266 |
try:
|
267 |
+
# İlk birkaç chunk için audio formatını analiz et
|
268 |
if self.total_chunks < 3:
|
|
|
269 |
if len(audio_chunk) >= 4:
|
|
|
270 |
import struct
|
271 |
try:
|
|
|
272 |
first_sample = struct.unpack('<h', audio_chunk[:2])[0]
|
273 |
log_info(f"🔊 Audio format check - Chunk #{self.total_chunks}: First sample={first_sample}, Size={len(audio_chunk)} bytes")
|
|
|
|
|
|
|
|
|
274 |
except:
|
275 |
log_warning("⚠️ Could not parse as Linear16")
|
|
|
|
|
|
|
|
|
|
|
|
|
276 |
|
277 |
+
# ✅ Final result geldiyse audio gönderme ama queue'yu kontrol et
|
278 |
+
if not self.final_result_received:
|
279 |
+
# Send audio to Deepgram
|
280 |
+
self.live_connection.send(audio_chunk)
|
281 |
+
|
282 |
+
self.total_chunks += 1
|
283 |
+
self.total_audio_bytes += len(audio_chunk)
|
284 |
+
|
285 |
+
# Log progress
|
286 |
+
if self.total_chunks % 50 == 0:
|
287 |
+
log_debug(f"📊 Listening... {self.total_chunks} chunks, {self.total_audio_bytes/1024:.1f}KB")
|
288 |
|
289 |
+
# ✅ Her zaman queue'yu kontrol et ve result'ları yield et
|
290 |
while True:
|
291 |
try:
|
292 |
result = self.responses_queue.get_nowait()
|
293 |
+
|
294 |
+
# Log for debugging
|
295 |
+
log_debug(f"🎯 Yielding result: is_final={result.is_final}, text='{result.text}'")
|
296 |
+
|
297 |
if result.is_final:
|
298 |
+
self.final_result_received = True
|
299 |
+
yield result # ✅ Final result'ı yield et
|
300 |
+
# Artık yeni audio kabul etme
|
301 |
+
elif not self.final_result_received:
|
302 |
+
# Final gelmeden önce interim'leri yield et
|
303 |
yield result
|
304 |
+
|
305 |
except queue.Empty:
|
306 |
break
|
307 |
+
|
308 |
except Exception as e:
|
309 |
log_error(f"❌ Error streaming audio", error=str(e))
|
310 |
self.is_streaming = False
|