fixes

requirements.txt

@@ -1,4 +1,5 @@
 faster-whisper
 pyannote.audio
 python-dotenv
+soundfile
 tqdm

transcriber/diarization/pyannote_engine.py

@@ -1,6 +1,7 @@
 import logging
 from dataclasses import dataclass
 
+import soundfile as sf
 import torch
 from pyannote.audio import Pipeline
 from pyannote.audio.pipelines.utils.hook import ProgressHook
@@ -47,6 +48,10 @@ class DiarizationEngine:
         """
         logger.info("Diarizing: %s", audio_path)
 
+        data, sample_rate = sf.read(audio_path, dtype="float32")
+        waveform = torch.from_numpy(data).unsqueeze(0)
+        audio_input = {"waveform": waveform, "sample_rate": sample_rate}
+
         kwargs = {}
         if min_speakers is not None:
             kwargs["min_speakers"] = min_speakers
@@ -54,7 +59,7 @@ class DiarizationEngine:
             kwargs["max_speakers"] = max_speakers
 
         with ProgressHook() as hook:
-            diarization = self._pipeline(audio_path, hook=hook, **kwargs)
+            diarization = self._pipeline(audio_input, hook=hook, **kwargs)
 
         turns = []
         for turn, speaker in diarization.exclusive_speaker_diarization: