Craig Whisper Transcription Script

Powershell script that uses OpenAI whisper model locally to transcribe, clean, and combine the multi-track audio output of Discord Craig.

Overview

transcribe.ps1 is a PowerShell script designed to automate the transcription, cleaning, and combination of multi-track audio files, specifically those generated by a Discord Craig bot. The script leverages OpenAI's Whisper model to perform high-quality speech-to-text transcription locally on a Windows machine.

This script is ideal for users who need to process multi-track audio recordings, such as podcasts, interviews, or collaborative discussions, and generate consolidated, easy-to-read transcripts in multiple formats.

Features

• Automated Transcription: Uses OpenAI's Whisper model (large-v2) locally for accurate transcription of audio files.
• Multi-Track Support: Processes multiple audio files with a naming convention (n-playername_m) to identify speakers.
• Error Handling: Logs errors and tracks transcription progress to allow resumption of interrupted processes.
• Output Formats:
  • Individual TSV files for each transcription.
  • Combined TSV file with all transcripts sorted by timestamp.
  • Plain text transcript (final_transcript.txt).
  • Markdown-formatted transcript (transcript.md) with timestamps and speaker names (ideal for building a custom GPT).
• Post-Processing:
  • Consolidates consecutive identical lines from the same speaker.
  • Filters out noisy words (e.g., "you") if they appear alone.
• Statistics: Collects and displays processing statistics, including total files processed, success/failure counts, and average processing time.
• Cleanup Option: Removes temporary files after processing if specified.

Requirements

To use the script, ensure the following dependencies are installed and available in your system's PATH:

• PowerShell 7.5.0
• Python (with pip installed)
• OpenAI Whisper (pip install -U openai-whisper)
• FFmpeg

The audio files must be in a format supported by FFmpeg (e.g., MP3, WAV, M4A, FLAC, OGG, AAC, MP4, WMA).

Usage

Run the script from a PowerShell terminal with the following parameters:

Parameters

• -InputFolder (Required): Path to the folder containing audio files to transcribe.
• -OutputFolder (Optional): Path to the folder where transcription outputs will be saved. Defaults to <InputFolder>\..\transcriptions.
• -Force (Optional): Forces re-transcription of already processed files.
• -PostProcessOnly (Optional): Skips transcription and only performs post-processing on existing transcripts.
• -Cleanup (Optional): Removes temporary files after processing.

Examples

1. Transcribe all audio files in a folder:

   .\transcribe.ps1 -InputFolder "C:\path\to\audio\files"