Initialer Push, Grundsätzlich funktioniert das Programm

2026-07-09 23:00:31 +02:00
parent 70088f7861
commit 952bbb6158
51 changed files with 3066 additions and 0 deletions
@@ -0,0 +1,65 @@
+import whisperx
+import gc
+import os
+import sys
+
+HF_TOKEN="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"
+DEVICE = "cuda"
+COMPUTE_TYPE = "float16"
+BATCH_SIZE = 4
+
+if len(sys.argv) < 2:
+    print("Verwendung: python transkribiere_datei.py <audiodatei> [...]")
+    sys.exit(1)
+
+audio_files = sys.argv[1:]
+print(f"{len(audio_files)} Datei(en) zu transkribieren\n")
+
+print("Lade Whisper large-v3...")
+model = whisperx.load_model("large-v3", DEVICE, compute_type=COMPUTE_TYPE, language="de")
+
+print("Lade Diarization-Pipeline...")
+diarize_model = whisperx.diarize.DiarizationPipeline(token=HF_TOKEN, device=DEVICE)
+
+for audio_path in audio_files:
+    base = os.path.splitext(audio_path)[0]
+    out_path = base + ".txt"
+    fname = os.path.basename(audio_path)
+
+    print(f"\n{'='*60}")
+    print(f"Verarbeite: {fname}")
+    print(f"{'='*60}")
+
+    audio = whisperx.load_audio(audio_path)
+
+    print("  1/4 Transkription...")
+    result = model.transcribe(audio, batch_size=BATCH_SIZE, language="de")
+
+    print("  2/4 Alignment...")
+    model_a, metadata = whisperx.load_align_model(language_code="de", device=DEVICE)
+    result = whisperx.align(result["segments"], model_a, metadata, audio, DEVICE,
+                            return_char_alignments=False)
+    del model_a; gc.collect()
+
+    print("  3/4 Speaker Diarization...")
+    diarize_segments = diarize_model(audio)
+    result = whisperx.assign_word_speakers(diarize_segments, result)
+
+    print("  4/4 Speichere Transkript...")
+    with open(out_path, "w", encoding="utf-8") as f:
+        f.write(f"Transkript: {fname}\n")
+        f.write("=" * 60 + "\n\n")
+        current_speaker = None
+        for seg in result["segments"]:
+            speaker = seg.get("speaker", "UNBEKANNT")
+            text = seg["text"].strip()
+            start = seg["start"]
+            end = seg["end"]
+            if speaker != current_speaker:
+                f.write(f"\n[{speaker}]\n")
+                current_speaker = speaker
+            f.write(f"  [{start:6.1f}s – {end:5.1f}s]  {text}\n")
+
+    print(f"  -> Gespeichert: {out_path}")
+
+print("\nAlle Dateien fertig!")