fix transcription in notebook

Whisper_Transcription_+_NeMo_Diarization.ipynb

@@ -812,15 +812,39 @@
         "# or run on CPU with INT8\n",
         "# compute_type = \"int8\"\n",
         "\n",
-        "whisper_results, language, audio_waveform = transcribe_batched(\n",
-        "    vocal_target,\n",
-        "    language,\n",
-        "    batch_size,\n",
-        "    whisper_model_name,\n",
-        "    compute_type,\n",
-        "    suppress_numerals,\n",
-        "    device,\n",
-        ")"
+        "whisper_model = faster_whisper.WhisperModel(\n",
+        "    whisper_model_name, device=device, compute_type=compute_type\n",
+        ")\n",
+        "whisper_pipeline = faster_whisper.BatchedInferencePipeline(whisper_model)\n",
+        "audio_waveform = faster_whisper.decode_audio(vocal_target)\n",
+        "suppress_tokens = (\n",
+        "    find_numeral_symbol_tokens(whisper_model.hf_tokenizer)\n",
+        "    if suppress_numerals\n",
+        "    else [-1]\n",
+        ")\n",
+        "\n",
+        "if batch_size > 0:\n",
+        "    transcript_segments, info = whisper_pipeline.transcribe(\n",
+        "        audio_waveform,\n",
+        "        language,\n",
+        "        suppress_tokens=suppress_tokens,\n",
+        "        batch_size=batch_size,\n",
+        "        without_timestamps=True,\n",
+        "    )\n",
+        "else:\n",
+        "    transcript_segments, info = whisper_model.transcribe(\n",
+        "        audio_waveform,\n",
+        "        language,\n",
+        "        suppress_tokens=suppress_tokens,\n",
+        "        without_timestamps=True,\n",
+        "        vad_filter=True,\n",
+        "    )\n",
+        "\n",
+        "full_transcript = \"\".join(segment.text for segment in transcript_segments)\n",
+        "\n",
+        "# clear gpu vram\n",
+        "del whisper_model, whisper_pipeline\n",
+        "torch.cuda.empty_cache()"
       ]
     },
     {