Hello everyone! I decided to blog on this topic because it reminds me of my hackathon project with my team. I attended my first-ever hackathon at UC Berkeley's AI Hackathon and developed "StoryTellerAI", an AI that tells stories to children. Integrating image-to-text and text-to-speech models was one of the hardest parts of our project. Hopefully, this blog will help those who need help on integrating image-to-text and text-to-speech models.

First, we need to learn the components of using this AI-generated model, so let's start with VLMs. Vision Language Models (VLMS) are a form of artificial intelligence that can understand and learn from visuals and linguistic modalities. A VLM can analyze an image or video and it will then generate a corresponding text description to the visual content.

Second, the next component of this AI model is TTS. Text-To-Speech (TTS) is the usage of human speech mimicry to read text aloud. This is done through phonemes -text broken down into the smallest units of sound. AI uses data on human speech patterns, tones, and rhythms to generate its own voice. This would allow the generated voice to have a personality instead of it sounding robotic. The system combines phonemes with AI to render a fully expressive speech output. Modern TTS systems are extremely advanced because of their remarkable capabilities in replicating different tones and voice inflections, working across languages, and understanding context.

These concepts are the backbone of TTS systems and its functionality is dependent on the client. TTS can be used to accomplish many things and will only continue to improve with research. Because its use is so diverse, TTS systems have a diverse library at their disposal and it can be altered however the user sees fit. The various implementations of TTS systems signify its flexibility and how it can be improved for future development.

This post was heavily inspired from:
Pambou, J. (2024, July 24) Integrating Image-To-Text And Text-To-Speech Models (Part 1)
Retrieved from: [https://www.smashingmagazine.com/2024/07/integrating-image-to-text-and-text-to-speech-models-part1/]