Chunk creation logic fails on Windows

[Lunat1q]

Scriberr/internal/transcription/adapters/py/nvidia/parakeet_transcribe_buffered.py

Line 94 in 850af1f

chunk_path = f"/tmp/chunk_{i}.wav"

Hi!
The line mentioned above fails for Windows machines.

time=22:13:39 level=ERROR msg="Job processing failed" worker_id=0 job_id=68d16f0f-e040-47c2-8f2c-1e95df65ccb0 error="single-track processing failed: transcription failed: Parakeet buffered execution failed: exit status 1\nLogs:\n0.0\r\n final_eos_score: 1.0\r\n score_per_phrase: 0.0\r\n source_lang: en\r\n use_triton: true\r\n uniform_weights: false\r\n use_bpe_dropout: false\r\n num_of_transcriptions: 5\r\n bpe_alpha: 0.3\r\n boosting_tree_alpha: 0.0\r\n hat_subtract_ilm: false\r\n hat_ilm_weight: 0.0\r\n max_symbols_per_step: 10\r\n blank_lm_score_mode: LM_WEIGHTED_FULL\r\n pruning_mode: LATE\r\n allow_cuda_graphs: true\r\n tsd_max_sym_exp: 50\r\n temperature: 1.0\r\n durations:\r\n - 0\r\n - 1\r\n - 2\r\n - 3\r\n - 4\r\n big_blank_durations: []\r\n \r\n✓ CUDA graphs disabled successfully\r\nSplitting audio into 300.0s chunks...\r\nCreated 12 chunks\r\nTranscribing chunk 1/12 (duration: 300.0s)...\r\nTraceback (most recent call last):\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\transcribe_buffered.py", line 171, in \r\n main()\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\transcribe_buffered.py", line 163, in main\r\n transcribe_buffered(\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\transcribe_buffered.py", line 84, in transcribe_buffered\r\n sf.write(chunk_path, chunk_info['audio'], sr)\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\.venv\Lib\site-packages\soundfile.py", line 363, in write\r\n with SoundFile(file, 'w', samplerate, channels,\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\.venv\Lib\site-packages\soundfile.py", line 690, in init\r\n self._file = self._open(file, mode_int, closefd)\r\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\r\n File "D:\AudioTranscribe\Scriberr_Windows_x86_64\data\whisperx-env\parakeet\.venv\Lib\site-packages\soundfile.py", line 1265, in _open\r\n raise LibsndfileError(err, prefix="Error opening {0!r}: ".format(self.name))\r\nsoundfile.LibsndfileError: Error opening '/tmp/chunk_0.wav': System error.\r\n"

I've modified it to be:

    temp_dir = "tmp"
    os.makedirs(temp_dir, exist_ok=True)
    chunk_path = os.path.join(temp_dir, f"chunk_{i}.wav")

so it creates the tmp folder right next to "exe" file. please validate that such approach will work with other OS.

[Lunat1q]

Also small addition for the Windows machine, before running the scriberr.exe from the console, I do run those 3

$env:SECURE_COOKIES="false" 
$env:PYTHONIOENCODING = "utf-8"
$env:UV_PYTHON = "3.12"

First one is optional if needed, but 2 and 3 are must, otherwise it flops.