Pipeline Overview

ssaTranscribe is an open-source, self-hosted pipeline that delivers audio-to-text speaker diarization, voice matching, multilingual vocabulary, and keyword detection in end-to-end transcriptions of episodic media.

Developed to support the expansion of the Sesame Street Archive (SSA), the primary goal of ssaTranscribe is to transcribe 4,397 original Sesame Street episodes aired between 1969-2018, enabling a multimodal approach to scaling the SSA from ~5,000 to 35,000 labeled images across face, number, place, and word object categories.

By surfacing rich dialogue from the world's most longstanding children’s educational television series, ssaTranscribe identifies visually imaginative, stylistically diverse scenes that reflect children's experiences of wonder, play, and embodied learning—strategically curating the magical content behind the SSA.

Directory Structure

This directory tree outlines the top-level folders and essential files included in ssaTranscribe:

ssaTranscribe/
├── data/                                        # Input and output data directories
│   ├── raw/                                     # Original unprocessed audio files
│   └── processed/                               # Audio clips segmented and ready for ASR
├── models/                                      # Pretrained model weights
│   └── asr_model.pt                             # Whisper or other ASR model file
├── scripts/                                     # Core logic for each pipeline step
│   ├── transcriptionwithspeakersv1.py           # Main script for processing audio to text
│   ├── diarize.py                               # Performs speaker diarization
│   └── postprocess.py                           # Formats and cleans transcripts
├── configs/                                     # YAML configuration files
│   └── default.yaml                             # Default settings for pipeline runs
├── README.md                                    # Project overview and usage
├── requirements.txt                             # Python package dependencies
└── run_pipeline.sh                              # End-to-end pipeline entry point

Data Flow

This table maps the flow of data through the ssaTranscribe pipeline, from original Sesame Street episode to readable, searchable output:

	Stage	Input Format	Process (File / Script)	Output Format	Output Location
1	Audio Ingestion	.wav, .mp3	Raw input in data/raw/	(same)	data/raw/
2	Segmentation / Prep	.wav, .mp3	scripts/segment.py (if used)	.wav (cleaned/trimmed)	data/processed/
3	Transcription (ASR)	.wav	scripts/transcribe.py using models/asr_model.pt	.txt, .json	outputs/transcripts/
4	Diarization (Optional)	.wav	scripts/diarize.py	.json speaker tags	outputs/transcripts/
5	Postprocessing	.json, .txt	scripts/postprocess.py	.srt, cleaned .json or .txt	outputs/transcripts/
6	Configuration	.yaml	configs/default.yaml	(used throughout)	—

Implementation

💪 Compute

Required GPU Type:

• .

Minimum CPU and RAM Specifications:

• .

Disk Space Requirements:

• .

Estimated Runtime Per Episode (With/Without GPU):

• .

---

📩 Installation

Python Version:

• .

Environment Setup Steps:

• .

Required Python Packages:

• .

Pretrained Model Download Instructions:

• .

System-Level Dependencies:

• .

---

⚙️ Configuration

Main Configuration File Location:

• .

User-Editable Parameters:

• .

Required Environment Variables:

• .

Command-Line Arguments, Flags:

• .

---

🎬 Execution

Full Pipeline Execution Command:

• .

Running Individual Pipeline Components:

• .

Expected Input and Output File Types:

• .

Default Output Directory and File Locations:

• .