March 2024 Summaries

This weekly update includes information about AssemblyAI's new Streaming Speech-to-Text product with updated end-of-utterance controls and lower pricing. The company has also released a free "Build an Audio AI App" course by Talk Python, which already has over 1k students. Additionally, the blog features tutorials on redacting PII from audio and video files using Python, business use cases for Generative AI, transcribing phone calls in real-time with Go, and more.

AssemblyAI's Streaming Speech-to-Text (previously Real-Time) model offers improved customization and control, making it easier to build next-generation AI tools on top of live speech transcription. The technology is now being used for various purposes such as live captions for streaming audio and video, voice assistants and bots for customer service, language learning tools, accessibility applications, and virtual meetings. Developers can customize end-of-utterance detection to create more natural interactions with their AI tools. Streaming Speech-to-Text allows users to transcribe live audio streams at a lower cost of $0.47 per hour or $0.0001306 per second, including access to additional features like Automatic Punctuation and Casing, and Custom Vocabulary.

Advanced Large Language Models (LLMs) are revolutionizing industries by enabling machines to create new content with machine learning models. These LLMs power chatbots, image generators, and software that can handle complex user requests and return near-human results. Generative AI has immense potential in various sectors, including healthcare, customer service, content creation, event marketing, and education. Businesses can leverage this technology to enhance operations, improve the customer journey, and drive innovative product development. Some examples of Generative AI include text generation, image synthesis, music composition, and data augmentation. LLMs are deep learning neural networks trained on massive amounts of data and fine-tuned for specific applications. LeMUR is a framework that allows companies to build Generative AI audio analysis tools more easily by leveraging different LLMs. Some industry use cases for Generative AI include facilitating quality patient care in telehealth, assisting businesses in analyzing customer interactions, streamlining live event marketing, and enhancing learning management systems.

This tutorial teaches how to redact Personal Identifiable Information (PII) from audio and video files using the AssemblyAI Python SDK. The process involves setting up an environment, transcribing the file, printing the redacted transcript, fetching the redacted audio file, and running the program. PII categories such as medical conditions, email addresses, and credit card numbers can be redacted from both audio/video files and their textual transcripts. The AssemblyAI API key is required for this process, which can be obtained for free.

This weekly update highlights new product features, tutorials, and community updates from the company. A free 4+ hour video course on building an audio AI app using Python is now available. Ruby SDK examples have been added to all documentation pages. The blog includes articles on extracting insights from customer calls with LLMs in Python, trends in Graph Neural Networks for 2024, and how to do Speech-to-Text with Go. Additionally, popular YouTube tutorials cover integrating LLM frameworks into AI apps, coding an AI voice bot from scratch, transcribing live phone calls with Python, and more.

AssemblyAI has released version 1.3.0 of its Go SDK, introducing new features such as word search in transcripts and the ability to permanently delete data associated with LeMUR requests. The update also includes improvements to existing functionality, including support for sending real-time audio samples as binary data and a type change for the LeMUR Context field. Users are encouraged to provide feedback or report issues through GitHub.

"Build An Audio AI App," a comprehensive video course by Talk Python, is now available for free on the Talk Python Training website and in Android and iOS mobile apps. The course teaches participants how to create an app that works with audio from podcasts, covering topics such as generating transcripts, building search engines, prompt engineering, summarizing content, creating conversational Q&A experiences, developing FastAPI applications, using MongoDB for data modeling, employing HTMX for modern user interfaces, and running background jobs. The course is led by Michael Kennedy, a professional software trainer and host of the popular Python podcast "Talk Python To Me." It's 100% free and can be accessed on the Talk Python platform. A Discord community is available for those seeking support or wishing to connect with other developers.

Large Language Models (LLMs) are revolutionizing how we interact with machines, automating tasks, and providing deep insights. They can generate content, solve complex problems, and mimic human-like conversations. LLMs have diverse applications such as audio data analysis, content creation, customer support, language translation, sentiment analysis, education, and cybersecurity. By leveraging these models, businesses can streamline operations, enhance productivity, and foster innovation without sacrificing their budget or other to-dos.

This week's update includes a significant update to LeMUR Docs, focusing on tutorials and prompt engineering guides. Four key use cases for LeMUR are highlighted: asking questions about audio data, summarizing audio data, extracting and generating data, and content generation based on audio data. Additionally, a new version of the Go SDK has been released with word search support and request data purging capabilities. The AssemblyAI blog features articles on extracting phone call insights with LLMs in Python, AI trends in 2024 focusing on Graph Neural Networks, and how to do Speech-to-Text with Go. Trending YouTube tutorials include coding an AI voice bot from scratch, transcribing live phone calls with Python - Flask tutorial, and implementing real-time speech-to-text transcription in Google Docs using LLMs (Python Tutorial).

The use of AI in business applications is rapidly growing, with AI expected to boost U.S. GDP by 2030. Generative AI tools like ChatGPT have been particularly popular. Companies across industries are integrating AI into their tools and platforms, but there are several key considerations product teams should address before building AI-powered tools for businesses. These include whether the integration helps consolidate or expand the tech stack, if they want to manage the AI model internally or have it managed externally, what kind of internal and external support is needed, security and data privacy needs, cost considerations, and ensuring that the AI model or LLM can grow with the company.

Market research platforms are leveraging AI models such as Speech-to-Text, Audio Intelligence, and Large Language Models (LLMs) to build advanced analysis tools for their customers. These AI models help transcribe voice and video feedback efficiently, generate key highlights and insights from customer responses, and categorize, tag, and search data. By using these technologies, market research platforms can offer more robust tools that provide valuable insights into customer behaviors and attitudes, helping companies make informed decisions based on customer feedback.

This week's update includes information about new product features, tutorials, and community updates from AssemblyAI. The Java SDK has been updated with improved real-time transcription controls, new content safety features, and the addition of Anthropic Claude 2.1 model to LeMUR. Additionally, there are blog posts on AI trends in 2024 focusing on Graph Neural Networks and tutorials on transcribing live phone calls using Python with Flask, indexing podcasts with keywords, and implementing real-time speech-to-text transcription in Google Docs using LLMs. To stay updated, sign up for AssemblyAI's newsletter.