FYL

A Google Colab pipeline that turns lecture audio into clean, structured Markdown notes.

FYL combines Faster-Whisper for speech-to-text and Gemini for note generation. It is designed for students who want a repeatable workflow: upload lecture audio to Google Drive, run the notebook in Colab, and receive a transcript plus structured notes in Markdown.

What it solves

• Transcribe long lecture recordings without building a local ASR setup
• Keep model cache and input/output files persistent through Google Drive
• Generate structured notes from raw transcripts using Gemini
• Support a low-cost workflow that works from any browser
• Keep the pipeline transparent: the notebook can be inspected, edited, and run step by step

Technical foundation

• Runtime: Google Colab
• Storage: Google Drive folders for input, output, and model cache
• Transcription: Faster-Whisper / CTranslate2
• AI notes: Gemini API through google-genai
• Output: Markdown notes saved in Drive
• Supported audio: .mp3, .wav, .flac

Workflow

flowchart LR
    A["Lecture audio"] --> B["Google Drive asr_in"]
    B --> C["Colab notebook"]
    C --> D["Faster-Whisper transcript"]
    D --> E["Gemini note generation"]
    E --> F["Markdown notes in asr_out"]

Loading

Quick start

1. Download FYL_gdrive-7.ipynb.
2. Upload it to Google Colab.
3. Create a Gemini API key from Google AI Studio.
4. Add the key to Colab Secret Manager.
5. Run the notebook once so it creates the Drive folders.
6. Upload audio files to:

MyDrive/asr_in

7. Run the notebook. Output files are written to:

MyDrive/asr_out

Recommended recording format

For reliable transcription and smaller uploads:

• Format: MP3
• Bitrate: 96 kbps
• Sample rate: 44.1 kHz
• Channels: mono

Audacity works well for converting raw recordings into this format.

Model cache size

Model	Approximate cache size
tiny	75-80 MB
base	140-160 MB
small	450-500 MB
medium	1.4-1.6 GB
large-v2	2.8-3.0 GB
large-v2-distilled	1.8-2.0 GB
large-v3	3.0-3.2 GB
large-v3-distilled	2.0-2.2 GB

Distilled models are smaller and faster, with a good tradeoff for students working inside limited Drive storage.

Why Colab and Drive

• Colab removes the need for a local GPU setup.
• Drive avoids slow repeated uploads and keeps the Whisper model cache persistent between runs.
• AI Studio gives students a direct path to Gemini API access.
• Markdown output stays portable and can be moved into Notion, Obsidian, Google Docs, or a personal knowledge base.

Project status

FYL is a notebook-first prototype. The core transcription and note-generation flow works, while future improvements may include reusable prompt presets, better chapter detection, and cleaner exports for different study workflows.

Responsible use

Use FYL with recordings you are allowed to process. If a class, institution, or speaker has rules about recording or AI transcription, follow those rules.

Credits

Built on top of Faster-Whisper and google-genai.