bump version 0.3.0

fix: cpu usage issue.

.github/workflows/ci.yml

@@ -77,7 +77,7 @@ jobs:
   build-and-push-docker-cpu:
     needs: [run-tests, check-code-format]
     runs-on: ubuntu-22.04
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags')
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
     steps:
       - uses: actions/checkout@v2
       
@@ -103,7 +103,7 @@ jobs:
     needs: [run-tests, check-code-format, build-and-push-docker-cpu]
     timeout-minutes: 20
     runs-on: ubuntu-22.04
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags')
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
     steps:
       - uses: actions/checkout@v2
 

docker/Dockerfile.gpu

@@ -30,4 +30,4 @@ COPY whisper_live /app/whisper_live
 
 COPY run_server.py /app
 
-CMD ["python", "run_server.py"]
+CMD ["python3", "run_server.py"]

whisper_live/__version__.py

@@ -1 +1 @@
-__version__ = "0.2.1"
+__version__ = "0.3.0"

whisper_live/server.py

@@ -408,6 +408,11 @@ class ServeClientBase(object):
         if self.frames_np is not None and self.frames_np.shape[0] > 45*self.RATE:
             self.frames_offset += 30.0
             self.frames_np = self.frames_np[int(30*self.RATE):]
+            # check timestamp offset(should be >= self.frame_offset)
+            # this basically means that there is no speech as timestamp offset hasnt updated
+            # and is less than frame_offset
+            if self.timestamp_offset < self.frames_offset:
+                self.timestamp_offset = self.frames_offset
         if self.frames_np is None:
             self.frames_np = frame_np.copy()
         else:
@@ -575,7 +580,7 @@ class ServeClientTensorRT(ServeClientBase):
             warmup_steps (int): Number of steps to warm up the model for.
         """
         logging.info("[INFO:] Warming up TensorRT engine..")
-        mel, _ = self.transcriber.log_mel_spectrogram("tests/jfk.flac")
+        mel, _ = self.transcriber.log_mel_spectrogram("assets/jfk.flac")
         for i in range(warmup_steps):
             self.transcriber.transcribe(mel)
 
@@ -613,7 +618,10 @@ class ServeClientTensorRT(ServeClientBase):
         """
         logging.info(f"[WhisperTensorRT:] Processing audio with duration: {input_bytes.shape[0] / self.RATE}")
         mel, duration = self.transcriber.log_mel_spectrogram(input_bytes)
-        last_segment = self.transcriber.transcribe(mel)
+        last_segment = self.transcriber.transcribe(
+            mel,
+            text_prefix=f"<|startoftranscript|><|{self.language}|><|{self.task}|><|notimestamps|>"
+        )
         if last_segment:
             self.handle_transcription_output(last_segment, duration)
 
@@ -793,7 +801,8 @@ class ServeClientFasterWhisper(ServeClientBase):
             task=self.task,
             vad_filter=self.use_vad,
             vad_parameters=self.vad_parameters if self.use_vad else None)
-        if self.language is None:
+
+        if self.language is None and info is not None:
             self.set_language(info)
         return result
 
@@ -878,7 +887,9 @@ class ServeClientFasterWhisper(ServeClientBase):
                 input_sample = input_bytes.copy()
                 result = self.transcribe_audio(input_sample)
 
-                if self.language is None:
+                if result is None or self.language is None:
+                    self.timestamp_offset += duration
+                    time.sleep(0.25)    # wait for voice activity, result is None when no voice activity
                     continue
                 self.handle_transcription_output(result, duration)
 
@@ -929,7 +940,6 @@ class ServeClientFasterWhisper(ServeClientBase):
         """
         offset = None
         self.current_out = ''
-        last_segment = None
         # process complete segments
         if len(segments) > 1:
             for i, s in enumerate(segments[:-1]):

whisper_live/transcriber.py

@@ -180,7 +180,7 @@ class WhisperModel:
 
         return config
 
-    def transcribe(
+    def transcribe(                                                         # noqa: C901
         self,
         audio: Union[str, BinaryIO, np.ndarray],
         language: Optional[str] = None,
@@ -315,6 +315,9 @@ class WhisperModel:
         else:
             speech_chunks = None
 
+        if audio.shape[0] == 0:
+            return None, None
+
         features = self.feature_extractor(audio)
 
         encoder_output = None