Class AMILoader

scripts/eval/eval.py:493–530

Dataset loader for AMI Meeting Corpus. The AMI corpus consists of 100 hours of meeting recordings captured using multiple microphones. This loader supports both Individual Headset Microphone (IHM) and Single Distant Microphone (SDM) conditions. Dataset Structure: - Text fil

class AMILoader(BaseDatasetLoader):
    """Dataset loader for AMI Meeting Corpus.

    The AMI corpus consists of 100 hours of meeting recordings captured using
    multiple microphones. This loader supports both Individual Headset Microphone (IHM)
    and Single Distant Microphone (SDM) conditions.

    Dataset Structure:
        - Text file with utterance IDs and transcripts
        - Audio files organized by meeting session
        - Multiple microphone configurations available

    Returns:
        Tuple of audio file paths and corresponding transcript texts

    Reference:
        Carletta, J., et al. "The AMI Meeting Corpus: A Pre-announcement."
    """

    def load(self) -> Tuple[list, list]:
        """Load AMI corpus audio files and transcripts.

        Parses the text file to extract utterance IDs and maps them to
        corresponding audio files in the evaluation subset.

        Returns:
            Tuple[list, list]: A tuple containing:
                - List of audio file paths (WAV format)
                - List of corresponding transcript strings
        """
        with open(f"{self.root_dir}/text", "r") as f:
            file_text = [line.split(" ", 1) for line in f]
            audio_files, transcript_texts = zip(*file_text)
            audio_files = [
                f"{self.root_dir}/{f.split('_')[1]}/eval_{f.lower()}.wav"
                for f in audio_files
            ]
        return list(audio_files), list(transcript_texts)


class CORAALLoader(BaseDatasetLoader):