smart-transcribe

What problem does it solve?

This skill tackles the tedious task of turning audio into accurate, searchable transcripts by orchestrating multiple transcription engines and a merging workflow that preserves speaker labels, builds a dictionary of corrections, and produces a final, production-friendly transcript with a transparency report.

Core Features & Use Cases

• Multi-engine transcription: run several engines (Cohere local, Mistral Voxtral, ElevenLabs Scribe v2, AssemblyAI Universal-3 Pro, and more) to improve accuracy and resilience.
• Agent merge workflow: automatically merge per-engine outputs via an agent-merge bundle, or headless merge modes for unattended runs.
• Dictionary-driven corrections: leverage a seed/user dictionary to correct common mishearings and proper-noun errors across transcripts.
• Per-context customization: context-specific terms and notes guide biasing and entity handling for meetings, interviews, and multi-speaker recordings.
• Output artifacts: produce final transcript markdown, an agent merge bundle, and a transparency report for review and auditing.

Quick Start

Transcribe a sample audio file using the default four-engine pipeline and generate an agent-merge bundle.