added api for ai; got backend working

transcribe.py

@@ -0,0 +1,91 @@
+#!/usr/bin/env python3
+import sys
+import json
+import tempfile
+import subprocess
+from faster_whisper import WhisperModel
+
+def extract_audio(input_path, output_path):
+    """Extract audio from video/audio file to 16kHz mono WAV"""
+    cmd = [
+        'ffmpeg', '-y', '-i', input_path, '-vn', '-ar', '16000', '-ac', '1', '-f', 'wav', output_path
+    ]
+    subprocess.run(cmd, check=True)
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: python transcribe.py <audio_file> <model_name> [language]", file=sys.stderr)
+        sys.exit(1)
+
+    audio_file = sys.argv[1]
+    model_name = sys.argv[2]
+    language = sys.argv[3] if len(sys.argv) > 3 else None
+
+    # Extract audio to temp WAV if needed
+    with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp:
+        wav_path = tmp.name
+
+    try:
+        extract_audio(audio_file, wav_path)
+
+        # Load model - use GPU if CUDA is available, else CPU with int8
+        import ctypes
+        try:
+            ctypes.CDLL("libcublas.so.12")
+            device = "cuda"
+            compute_type = "float16"
+        except OSError:
+            device = "cpu"
+            compute_type = "int8"
+
+        model = WhisperModel(model_name, device=device, compute_type=compute_type)
+
+        # Transcribe
+        segments, info = model.transcribe(
+            wav_path,
+            language=language,
+            beam_size=5,
+            word_timestamps=True,
+            vad_filter=True,
+            vad_parameters=dict(threshold=0.5, min_speech_duration_ms=250),
+            without_timestamps=False
+        )
+
+        # Convert to our format
+        words = []
+        segments_list = []
+
+        for segment in segments:
+            seg_words = []
+            for word in segment.words:
+                w = {
+                    "word": word.word,
+                    "start": word.start,
+                    "end": word.end,
+                    "confidence": word.probability
+                }
+                words.append(w)
+                seg_words.append(w)
+
+            segments_list.append({
+                "id": len(segments_list),
+                "start": segment.start,
+                "end": segment.end,
+                "text": segment.text,
+                "words": seg_words
+            })
+
+        result = {
+            "words": words,
+            "segments": segments_list,
+            "language": info.language
+        }
+
+        print(json.dumps(result))
+
+    finally:
+        import os
+        os.unlink(wav_path)
+
+if __name__ == "__main__":
+    main()