May 2024 Summaries

This week's update includes information about new product features, tutorials, and community updates from AssemblyAI. Users can learn how to use Large Language Models (LLMs) with LeMUR for various tasks such as asking questions, summarizing, extracting, and generating content from audio data. Additionally, there are blog posts on using Node.js for speech-to-text transcription with punctuation, casing, and formatting, filtering profanity from audio files using Node.js, and content moderation on audio files with Python. The latest YouTube tutorials cover real-time speech to text in Java, testing LLMs with audio data using Postman, and building a talking AI with LLAMA 3 (Python tutorial). Users can sign up for updates on new product features and weekly tutorials & blogs.

The process of transcribing audio to text has significantly improved due to the evolution of Speech AI models, which provide high accuracy rates. Transcription is now a non-negotiable tool for thousands of businesses across various industries. Accurate transcriptions allow users to create searchable archives, summarize key takeaways, find specific topics or information, maintain compliance, and unlock valuable insights. However, accurately transcribing audio at scale can be challenging due to factors such as data volume, audio quality, diverse accents and dialects, and background noise. Advanced speech recognition technology and cutting-edge AI models now enable anyone to transcribe audio to text at virtually any scale. The benefits of using Speech AI for transcription include making data searchable, improving accessibility, simplifying analysis, boosting compliance, increasing insights, and streamlining workflows. To get started with AssemblyAI's transcription services, users need to install and configure the SDK, submit their audio file, and enable additional AI models if desired. Key considerations when choosing a Speech AI provider include high accuracy with advanced AI models, support for multiple languages, understanding noisy data, speaker diarization, streaming speech-to-text transcription, cloud-based scalability, cost savings, and integration with business systems.

This tutorial demonstrates how to add punctuation, casing, and formatting to automatically-generated transcripts using the AssemblyAI JavaScript SDK. The process involves setting up an environment with Node.js, installing the AssemblyAI SDK, obtaining an API key, transcribing audio files with specific options (e.g., punctuate: true, format_text: true), and printing the filtered text. Additional configuration options and information about transcript objects are available in the AssemblyAI docs.

This tutorial teaches how to use Node.js to filter profanity from audio files using the AssemblyAI JavaScript SDK. First, set up your environment by installing Node.js 18 or higher and creating a new project folder. Then, install the AssemblyAI package and configure an API key as an environment variable. Next, transcribe the audio file with profanity filtering enabled and print the filtered text to the terminal. The transcript contains additional information such as word-level timestamps.

In this tutorial, we learn how to use Python and state-of-the-art AI models for content moderation of audio files. We set up the environment by installing Python and the assemblyai package, then transcribe an audio file with content moderation enabled using aai.Transcriber object. The resulting transcript contains information about any potentially sensitive segments in the file. We print the content moderation results to identify sections of audio that are considered sensitive according to some internal criterion. Finally, we summarize the overall findings of the Content Moderation model and get a finer-grained breakdown of these issues by accessing severity score summary.

Companies are increasingly integrating Speech AI technology to turn voice data into valuable insights, revolutionizing healthcare market research techniques. By leveraging AI, businesses like GRG Health can enhance traditional methods of gathering and analyzing healthcare market insights by streamlining data processing for faster and more accurate results. Some key ways individuals and companies use Speech AI in healthcare market research include improving data analysis through searchable audio files, organizing, categorizing, and annotating for deeper insights, understanding the patient experience better, and securing sensitive information with features like PII redaction.

This week's update includes information about new language support for Speaker Diarization, fresh blog posts, and trending YouTube tutorials from AssemblyAI. The Speaker Diarization model now supports five additional languages: Chinese, Hindi, Japanese, Korean, and Vietnamese. Blog posts cover topics such as automatically determining video sections with AI using Python, filtering profanity from audio files using Python, and how to use audio data in LlamaIndex with Python. The YouTube tutorials focus on building an AI voice bot from scratch, testing LLMs with audio data, and creating a talking AI using LLAMA 3.

This tutorial teaches how to use Python to filter profanity from audio files using the AssemblyAI API. First, install the assemblyai package and set up an environment variable for your API key. Next, transcribe the audio file with profanity filtering enabled by creating a TranscriptionConfig object and passing it into a Transcriber object. Finally, print the filtered text. The transcript contains additional information like word-level timestamps which can be accessed through the object's attributes.

This week's update includes information about new features in AssemblyAI's Speech-to-Text API, such as Speaker Diarization and PII Redaction. The latest benchmarks for the API are also highlighted. Additionally, the release of AssemblyAI JavaScript SDK 4.4.3 is announced with documentation and an API reference provided. New blog posts cover topics like automatically determining video sections with AI using Python, redacting PII in audio with Make and AssemblyAI, and how to use audio data in LlamaIndex with Python. Trending YouTube tutorials include testing LLMs with audio data using Postman, building a talking AI with LLAMA 3, and improving user experience with customizable real-time speech-to-text.

Speech AI technology is being utilized in various innovative ways across industries, from enhancing educational tools to improving customer support. Some of the notable use cases include meeting note takers and co-pilots, tutoring systems that adapt to individual learning styles, digital advertisement protection, automated data analysis, call analysis and conversation intelligence, video editing, hiring assistants, voice overs, gaming chatbots, accessibility tools, mental health monitoring, real-time translation, autonomous retail assistants, elderly care communication assistance, emergency dispatch, sales coaching, voice ordering bots, and media and archive search. These applications demonstrate the immense potential of Speech AI technology in transforming various aspects of daily life and work.

This week's update introduces new Best and Nano tiers for Speech-to-Text models, offering users more flexibility in balancing accuracy and cost. The Best tier uses the Universal 1 model to provide high accuracy where every detail matters, while the Nano tier offers high-quality transcripts at an affordable price. Users can select the tier based on their specific needs to build applications tailored to their requirements and budget constraints. Additionally, AssemblyAI's architecture is discussed in a deep dive for the 'This Is My Architecture' series at AWS. The latest YouTube tutorials cover building better user experiences with customizable real-time Speech-to-Text, creating a talking AI using LLAMA 3 and ElevenLabs, and transcribing live phone calls with Python - Flask. Blog articles discuss the use of AI in product teams for call tracking solutions, top considerations before building with AI models and LLMs, and how to use AI to build powerful market research tools.

AssemblyAI has introduced two new Speech-to-Text tiers, Best and Nano, offering different levels of power, capabilities, and price points. The Best tier is the default for customers and provides highly accurate transcriptions suitable for complex audio files with noisy backgrounds, multiple speakers, or accented speech. The Nano tier offers high-quality Speech-to-Text at an accessible price point, ideal for content generation, topic detection, and testing out models in a low-cost way. Users can easily switch between tiers based on their specific needs and budget constraints.

This week's update includes a new interactive API reference for testing APIs in real-time, expanded codec support, and bug fixes. Additionally, there are three new blog posts on redacting PII in audio with Make and AssemblyAI, detecting scam calls using Go with LeMUR and Twilio, and transcribing audio and video files with Python and Universal-1. Furthermore, the latest YouTube tutorials cover building a server-to-server app that transcribes Zoom recordings, creating an AI lecture assistant with Python, and speech recognition in Java.