Tag: multispeaker transcription

  • Linguistic Interview Transcriber for Turkish

    In linguistic and sociolinguistic research, transcribing recorded interviews can be a time-consuming part of the research process, particularly when recordings involve multiple speakers. Linguistic Interview Transcriber for Turkish is an open-source tool designed to automate the transcription of Turkish linguistic interviews using automatic speech recognition and speaker diarization.

    The tool uses “WhisperX” with the “large-v3-turbo” model to produce time-aligned transcripts and automatically distinguish between speakers in the recording. The model was selected primarily for its faster inference, making the tool more practical for processing longer interview recordings.

    My aim is to provide researchers with a simple and reusable transcription workflow that can be applied to Turkish interview recordings without requiring manual transcription from scratch.

    What the tool does:

    • Transcribes Turkish speech automatically
    • Provides time-aligned transcription
    • Performs speaker diarization to separate different speakers
    • Supports interviews with multiple speakers
    • Produces transcripts in several common formats, including TXT, JSON, SRT, VTT, and TSV
    • Preserves Turkish characters and timestamps
    • Uses for “large-v3-turbo” faster transcription
    • Can be used as a starting point for linguistic and sociolinguistic annotation

    For interviews involving three or four speakers, speaker diarization can be particularly useful for keeping speakers’ contributions separated. If the expected number of speakers is known, specifying it is recommended. However, diarization performance can vary depending on recording quality, microphone configuration, overlapping speech, background noise, and similarities between speakers’ voices. Speaker labels should therefore be checked before being used for linguistic analysis.

    Usage:

    1. Install the required dependencies
    2. Provide the path to the interview recording
    3. Run the transcription script
    4. Check the generated transcript and speaker labels
    5. Use the output as a starting point for further linguistic analysis or annotation

    GitHub Repository:

    https://github.com/bostanberkay/linguistic-interview-transcriber-for-turkish

    The tool is intended to be reusable by researchers working with Turkish linguistic and sociolinguistic interview data. I personally used this code for my own TUBITAK Project for transcription.

    You can always contact me for further issues, suggestions, or improvements!!