Initial commit

transcriber.py

@@ -0,0 +1,60 @@
+from typing import Any
+
+import whisper
+
+from utils import TimelineEvent
+
+class Transcriber:
+    """This class performs transcription of the audio file."""
+    @staticmethod
+    def get_models_list() -> list[str]:
+        """Returns list of allowed `model` values for constructor."""
+        return whisper.available_models()
+
+    def __init__(self, model: str, **kwargs) -> None:
+        """Create the transcriber instance.
+        
+        Args:
+            - model - model to use (call `get_models_list` to get the list of
+                      available models)
+            - **kwargs are passed to `whisper.load_model(...)`
+        """
+        allowed_models = self.get_models_list()
+        if model not in allowed_models:
+            raise ValueError(
+                f"Model `{model}` is not available. "
+                f"Use one of {', '.join(f'`{s}`' for s in allowed_models)}"
+            )
+        self._model = whisper.load_model(model, **kwargs)
+
+    def transcribe(self, path: str, **kwargs) -> list[TimelineEvent]:
+        """Transcribe audiofile. The operation will take a lot of time for large
+        files.
+
+        Args:
+            - path - path to the file to transcribe.
+            - **kwargs - passed to `transcribe()`
+        
+        Returns:
+            - list of timeline events you should use
+        """
+        raw_segments: list[dict]
+        raw_segments = self._model.transcribe(path, **kwargs)["segments"] # type: ignore
+        result: list[TimelineEvent] = []
+        seg_id: int = 0
+        for raw_segment in raw_segments:
+            ev = TimelineEvent(
+                id = f"asr_{seg_id}",
+                timestamp=float(raw_segment["start"]),
+                duration=float(raw_segment["end"]) - float(raw_segment["start"]),
+                payload=raw_segment["text"],
+                custom={
+                    "whisper_temperature": float(raw_segment["temperature"]),
+                    "whisper_avg_logprob": float(raw_segment["avg_logprob"]),
+                    "whisper_no_speech_prob": float(raw_segment["no_speech_prob"])
+                },
+                links=[]
+            )
+            result.append(ev)
+            seg_id += 1
+        return result