youtube-transcript-pipeline

What This Skill Does

The youtube-transcript-pipeline is a sophisticated automation workflow designed to transform raw YouTube audio into high-fidelity, structured transcript deliverables. It handles the complete lifecycle of audio processing, from initial download and chunked transcription via Deepgram's Nova-3 model to multi-stage post-processing. Key capabilities include speaker diarization correction, optional bilingual translation (English-to-Chinese), and professional-grade packaging into standardized project directories. By maintaining strict adherence to timestamp and line-format integrity, the skill ensures that all intermediate artifacts—raw transcripts, cleaned versions, and translated snippets—remain auditable and consistent for downstream editorial or research use.

Installation

To install this skill, use the OpenClaw CLI in your terminal: clawhub install openclaw/skills/skills/bluebirdback/youtube-transcript-pipeline Ensure you have your Deepgram API key configured in your environment variables, as the pipeline relies on the nova-3 model for high-accuracy diarization and word-level timing.

Use Cases

• Journalistic Interviews: Transcribing long-form interviews with clear speaker separation between the journalist and the guest.
• Content Repurposing: Automatically generating cleaned, structured transcript files for blog posts, social media, or archival purposes.
• Cross-Lingual Research: Translating technical interview transcripts into Chinese while preserving original timestamps and speaker tags to maintain a clear link back to the source audio.
• Content Auditing: Maintaining a repository of raw, cleaned, and translated transcripts for complex editorial projects where version control is critical.

Example Prompts

1. "Use the youtube-transcript-pipeline to transcribe this interview: [URL]. Please focus on cleaning up the speaker attribution so the interviewer is identified as 'Host' and the interviewee as 'Guest'."
2. "Download the audio for [URL], generate a transcript with Deepgram, and then provide a Chinese translation of the resulting text while keeping the exact timestamp format."
3. "Package the latest transcription project into a standard workbench folder. Ensure the MANIFEST.txt is updated and include both the raw and the speaker-corrected versions."

Tips & Limitations

• Speaker Attribution: Always perform a conservative correction pass first. The pipeline is designed to be cautious to prevent over-attributing short interjections to the wrong speaker.
• Audio Quality: While Nova-3 is powerful, ensure the input audio is relatively clean; heavy background noise may degrade speaker identification accuracy.
• Versioning: Use the scripts/ folder provided to handle workflow helpers. It is highly recommended to run the create_youtube_transcript_workbench.sh script initially to establish the correct directory hierarchy, as the pipeline expects this structure for automated packaging and manifest generation.