Our Token Methodology

A token is a discrete, reusable piece of data that AI models use to process information. Tokens can come from text, audio, images, or video.

• Text tokens: Words, parts of words, or punctuation. Example: “The purple squirrel runs” → “The,” “purple,” “squirrel,” “runs.”
  
• Audio tokens: Phonemes, sounds, or speech segments.
  
• Video/Image tokens: Frames, objects, visual features, or patterns.

Tokens are the fundamental building blocks of AI understanding. More high-quality tokens generally mean more sophisticated and accurate AI outputs.

Tokens are used by AI models, data scientists, and enterprises to train algorithms, improve search and analytics, and derive insights from unstructured data.

At Veritone:

• AI engines within aiWARE™ consume tokens to learn patterns and extract meaning.
  
• Enterprises can use tokens to automate workflows, make data-driven decisions, and uncover revenue opportunities through licensing or AI applications.
  

Content creators, broadcasters, and brands use tokens to index, search, and monetize media assets.

Tokens transform raw, unstructured data into structured, machine-readable, actionable information. Without tokens:

• AI models can’t efficiently process complex audiovisual content.
  
• Valuable information remains locked in unstructured formats like videos, podcasts, or images.
  
• Enterprises are data-rich but insight-poor, unable to leverage their assets fully.
  

Tokens allow data to be indexed, searched, monetized, or used as training material, enabling smarter AI and new business opportunities.

Veritone has been a token factory since its founding – we just didn’t call it that. Our internal schema forms the foundation of tokenizing unstructured audio and video. The process involves:

1. Segmentation: breaking raw audiovisual streams into structured events (frames, clips, speech segments).
   
2. Annotation: adding metadata such as speaker identity, location, timecodes, entities, and context.
   
3. Normalization: converting annotations into a consistent, machine-readable namespace so they are interoperable across aiWARE models and applications.
   

Each normalized metadata object becomes a token that can be indexed, searched, monetized, or used for AI training.

Veritone generates a rich set of metadata tokens, including:

• Identity tokens: e.g., speaker “Ryan Steelberg,” character “Robin Hood.”
  
• Time/location tokens: e.g., start/end timecodes, GPS coordinates, stadium names.
  
• Event tokens: e.g., goal scored, breaking news alert, foul committed.
  
• Semantic tokens: e.g., tone: “angry,” classification: “financial news,” object: “aircraft.”
  

Together, these tokens form a semantic ledger of meaning that can feed AI models, search systems, or even become a monetizable data asset.

Processing trillions of tokens demonstrates Veritone’s ability to:

• Transform massive amounts of unstructured data into AI-ready assets.
  
• Support multimodal AI, where text, audio, and video are processed together.
  
• Address enterprise data challenges, turning unused data into insights and revenue streams.
  

Establish Veritone and its products as the foundation for AI data monetization, from model training to licensing and financialized media assets.

Tokens are the lifeblood of AI, enabling models to learn, understand, and act on data. For enterprises, tokenization can:

• Improve operational efficiency through AI workflows and automation.
  
• Enable smarter decision-making by transforming raw media into structured data.
  
• Unlock new revenue opportunities through licensing, AI model training, or content monetization.
  

Our ability to process large volumes of tokens underscores Veritone’s leadership in converting unstructured data into actionable, high-value AI assets.