Initial commit

.gitignore

@@ -0,0 +1,4 @@
+dist/
+__pycache__/
+build/
+venv/

main.py

@@ -0,0 +1,73 @@
+import logging
+import os
+import warnings
+import wave
+from datetime import datetime
+
+import pyaudio
+import whisper
+
+logging.basicConfig(level=logging.DEBUG)
+warnings.filterwarnings('ignore')
+
+model = whisper.load_model("tiny")
+
+frames_per_buffer = 1024
+sample_format = pyaudio.paInt16
+channels = 1
+rate = 44100
+clip_duration = 4
+model_seconds = 5
+overlap = 0
+
+p = pyaudio.PyAudio()
+
+logging.debug("Recording...")
+
+frames = []
+
+
+def stream_callback(in_data, frame_count, time_info, status):
+    frames.append(in_data)
+    return in_data, pyaudio.paContinue
+
+
+stream = p.open(format=sample_format,
+                channels=channels,
+                rate=rate,
+                frames_per_buffer=frames_per_buffer,
+                input=True,
+                stream_callback=stream_callback)
+
+stream.start_stream()
+
+fps = int(rate / frames_per_buffer * clip_duration)
+
+while True:
+    if len(frames) > fps:
+        fname = ''.join(
+            ['.', '/clip-', datetime.utcnow().strftime('%Y%m%d%H%M%S'), '.wav'])
+        try:
+            clip = []
+            for i in range(0, fps):
+                clip.append(frames[i])
+
+            wavefile = wave.open(fname, 'wb')
+            wavefile.setnchannels(channels)
+            wavefile.setsampwidth(p.get_sample_size(pyaudio.paInt16))
+            wavefile.setframerate(rate)
+            wavefile.writeframes(b''.join(clip))
+            wavefile.close()
+
+            result = model.transcribe(audio=fname, language="en")
+            print(result["text"])
+
+            frames = frames[fps:]
+            logging.debug("Buffer size: ", len(frames))
+        except KeyboardInterrupt as e:
+            logging.debug("Ending recording...")
+            stream.stop_stream()
+            stream.close()
+            p.terminate()
+        finally:
+            os.remove(fname)