March 2022 Summaries

The text discusses the gender gap in Automatic Speech Recognition (ASR) technology and how it intersects with women's voices and speech patterns. It highlights that while English has relatively little overt gender marking, there are distinct vocal characteristics associated with women's speech. Some commentators see female speech as a tool for social connection and power, while others criticize certain speech patterns like uptalk and vocal fry. The text also explores the role of technology in this problem, noting that early decisions in voice recording limited the "voiceband" range to better serve male voices. Deepgram's ASR data shows a very small accuracy gap between male and female speakers, which is an encouraging sign for gender equality in speech technology.

This article discusses upgrading the freeCodeCamp Quote Generator project by using Deepgram's Node SDK to fetch a quote. The author provides detailed steps on how to prepare the project, explains the basic functionality of the original quote generator, and outlines the new approach that involves transcribing audio files from a JSON file of movie quotes using Deepgram's Node SDK. The article also covers setting up the necessary packages for the project, connecting the app.js file to the server.js file, and testing the updated project in a local environment.

The text discusses using Python and speech recognition technology for monitoring script compliance during phone conversations. It mentions a common use case for this is in call centers, where agents are required to inform customers that their calls are being recorded. Deepgram's speech-to-text solution can be used with Python to transcribe audio to text, enabling analytics and improving customer experience. The article then provides a step-by-step guide on how to build a simple script using the Deepgram Python SDK to monitor search terms in an audio file.

Deepgram has expanded its language offerings to over 20 models, including various dialects. This move demonstrates the company's commitment to providing accurate speech recognition solutions for businesses with international locations or customers. The new suite of languages is a significant step towards delivering a global language experience on par with their English model. Deepgram's end-to-end deep learning architecture and transfer learning processes enable them to offer the most cost-effective, fast, and accurate speech-to-text system in an ever-expanding number of languages. These new language models are now available for developers at no extra charge.

Technical writing aims to be understood by a general audience, including non-native English speakers and beginners in the field. To make technical writing accessible, writers should avoid jargon, acronyms, and numeronyms, and use clear language and structure. Breaking down content into paragraphs with headings and subheadings can also improve accessibility. Giving clear directions, using meaningful link text, and assessing the writing for clarity are essential steps in creating accessible technical content. Resources such as WC3, ASE Simplified Technical English, and Amy Dickens' talk on accessible language provide further guidance on this topic.

This article discusses how to write Vue 3 composables for integrating a third-party API. It explains the concept of composables, which are functions that encapsulate stateful logic and can be reused throughout a project. The author demonstrates how to create two composables: one to bring in an external script (useIVSPlayer) and another to connect to an Amazon IVS streaming channel (useIVSChannel). The article also covers the organization of files, setting up the video player component, and running the composables. It concludes by mentioning that future posts will cover using Deepgram's API for creating text captions and other related topics.

ARTiculate is a hands-free drawing application that utilizes voice commands for artistic expression, aiming to increase accessibility for people with disabilities. Inspired by VoiceDraw, the team behind ARTiculate developed advanced features such as voice-controlled color mixing and shortcuts for navigating the canvas. The project was built using Deepgram's Speech Recognition API and P5.js library for creative coding in JavaScript. Future extensions include image importing from online sources and additional accessibility options like custom voice commands and color-blindness support.

This article explores how to combine speech recognition with Python using Deepgram and speech-to-text analytics. The tutorial demonstrates building a simple script that does voice transcription with Python, utilizing the diarize feature of Deepgram to recognize multiple speakers. Analytics are then focused on measuring the amount of time each speaker spoke per phrase, the average amount of time they spoke, and the total time of conversation for each speaker. The code is provided along with instructions on how to run the project.

Deepgram has recently been recognized as the number one Voice Recognition Software on G2 and received a Silver Stevie Award for Customer Service Solutions Technology Partner of the Year. The company's focus on customer service and success is a driving force behind these recognitions. Deepgram aims to build partnerships with its customers, ensuring they achieve their desired outcomes and see value in their partnership. The company provides dedicated technical resources and customer success managers to advocate for each customer throughout their journey.

In this tutorial, a voice-controlled game is created using P5.js, a creative coding library that simplifies working with the Canvas API. The game features an enemy appearing from one of four directions every few seconds and moving towards the player. If the player says a specific word associated with each direction correctly, a bullet will fly in that direction. The final code for this project can be found on GitHub. To create the game, various P5 functions are used such as preload(), setup(), draw(), and keyPressed(). Additionally, Deepgram's Speech Recognition API is integrated to enable voice control of the game.

Technical writing involves a balance of ethos (credibility), pathos (emotion), and logos (logic). Ethics in technical writing means being honest, accurately representing evidence, and giving credit to sources. To ensure ethical writing, evaluate information included, words used, potential pressures, and whether credit is given to resources. Avoid deceptive writing by not blurring lines or misrepresenting data. Give credit to referenced authors through links, footnotes, or citations. Unintentional ethics violations can occur due to misinformation, biases, inaccurate information, and lack of consideration for the primary audience. Regularly verify information and update content as needed.

A speech-to-text model for the Ukrainian language has been released by Deepgram to assist with humanitarian efforts in Eastern Europe. The model is designed to transcribe Ukrainian and can be used in call centers set up to help refugees fleeing the conflict. Access to this model will be free for at least six months. Users can sign up for a Deepgram account, select Ukrainian as their audio language, and use the API by adding "language=uk" to transcription requests.

In this tutorial, the author guides readers through building a live streaming web application using Vue and Express setup. The full-stack application includes features such as a landing page that asks for a code to enter the site, a router (Vue Router) to implement routes and a navigation guard, a state management library (Vuex), and a backend server with two endpoints. The author also discusses setting up Vue Router and Vuex, creating a protected entry page with a navigation guard, and setting up the backend server for security purposes.

Triolingo is a chatbot designed by a team of developers at Hack Cambridge to help language learners improve their conversation skills through practice with a bot. The chatbot uses GPT-3 from OpenAI for dynamic responses and Deepgram's Python SDK for speech recognition, supporting multiple foreign languages. Users can select a conversation topic and target language, and the bot will respond with relevant prompts. Triolingo aims to fill the gap left by online language learning platforms that do not offer practice for speaking and listening skills. The team plans to incorporate additional Deepgram functionality and contextual challenges in future development.

The text discusses six challenges faced while developing an automatic speech recognition (ASR) model for Hindi. Firstly, Hindi is a low-resource language with limited data available for training ASR models compared to languages like English. Secondly, multilingualism and dialects pose difficulties as speakers may switch between languages or use loanwords from other languages, making it challenging for the model to track conversations accurately.

TomScottPlus is a Chrome extension that provides contextual overlays with Wikipedia links for YouTube videos created by English YouTuber Tom Scott, who covers topics such as geography, history, science, technology, and linguistics. The team behind the project was inspired by the need to make learning from these videos more seamless. The extension uses Deepgram's real-time audio analysis capabilities to identify relevant Wikipedia articles for each video topic. It then displays clickable overlays with links to those articles during playback. The source code is available on GitHub, and feedback or suggestions are welcome in the project's GitHub discussions.

This tutorial demonstrates how to integrate Deepgram's automatic speech recognition (ASR) engine into a Unity game for more immersive and accessible experiences. The tutorial assumes no prior experience with Unity or C#, but covers the process of setting up a simple physics simulation in Unity that uses Deepgram's ASR to trigger actions. It also provides ideas for building new features and enhancing games using speech recognition technology.

Technical Writing can be enhanced by incorporating storytelling techniques, making the content more memorable and engaging for readers. Storytelling helps create personal connections with audiences, allowing them to better understand concepts and relate to each other. Effective storytelling in technical writing involves knowing your audience, understanding the purpose of your writing, and implementing various strategies such as hooks, examples, response elements, gamification, theme, and visuals. By using storytelling effectively, technical writers can improve their readers' comprehension and retention of the content.

In this series, the author demonstrates how to build a live streaming web application with text captions using Amazon IVS and Deepgram. The main technologies used are Open Broadcast Software (OBS), Amazon Interactive Video Service (IVS), and Deepgram's speech-to-text API. Two versions of the project are built: one using vanilla HTML/Javascript, focusing on creating a front-end with limited back-end; and another using Vue.js/Node.js for added functionality and security features. The author provides step-by-step instructions for setting up OBS, Amazon IVS, and Deepgram, as well as code examples for integrating these technologies into the web application.

Automatic Speech Recognition (ASR) can help educators improve classroom accessibility for students with diverse backgrounds and needs. Current solutions such as posted slides, lecture recordings, and in-class notetakers have limitations. ASR technology offers a new solution by providing live, automatically generated transcripts during lectures. This method benefits students who struggle to hear the speaker, keep track of spoken content, or are non-native speakers. The live transcripts can be used on the fly and reviewed later for better understanding and organization of lecture material.

Deepgram has developed a unique 404 error page that not only provides helpful information but also offers an enjoyable gaming experience. The idea for this innovative approach originated from their internal hack-a-thons, where they brainstormed ideas and gathered feedback. To improve the user experience, Deepgram integrated a search function on the 404 page to help users find relevant content quickly. Additionally, they developed a game called MadGab, which is now available on their 404 page as an enjoyable distraction for users who encounter errors while navigating the site. The company's commitment to providing better experiences for developers has led them to focus on improving various aspects of user interaction with Deepgram, including sign-up processes, working in consoles, and using SDKs.

In this tutorial, learn how to create a speech-enhanced game using the Godot game engine and Deepgram's automatic speech recognition (ASR) engine. The game features a player who can move around and cast fireball spells by chanting "fire" into their microphone. The integration with Deepgram allows for nontraditional input devices, making games more accessible to players. This tutorial covers setting up the project, creating a player, spawning fireballs, and triggering the fireballs with voice commands using Deepgram's ASR engine.

Bias in automated speech recognition (ASR) systems has become an important issue for companies as it can impact customers negatively based on factors such as gender, race, age, and accent. Detecting bias in ASR systems involves identifying if transcripts for certain speakers have significantly higher word error rates than others. To address this issue, understanding the source of bias is crucial, which often comes from biased data used to train models. Solutions may include gathering more diverse data or adjusting the model's parameters to better handle edge cases and reduce bias.

The team behind yack! has developed an automatic video-to-comic book generator using Deepgram's Speech Recognition API and computer vision. The process involves generating a transcript with Deepgram, selecting keyframes in the video, applying comic book styling to the images, overlaying captions as speech bubbles, and placing each 'tile' in a dynamic SVG element. The project leverages Deepgram's utterances feature for understanding keyframes and diarization for color-coded text when different speakers are detected.

This tutorial introduces P5.js, a JavaScript library that simplifies working with the Canvas API for creating interactive and animated graphics. The article covers basic concepts such as setup() and draw(), color definitions, drawing shapes, setting fills and strokes, moving elements, handling user input through mouse and keyboard, and displaying images. It also provides examples of code snippets to demonstrate these concepts.

In this tutorial, we learn how to create a live speech-to-text transcription application using Python and Quart, an asynchronous microframework for building web applications. We also use Deepgram's AI speech recognition capabilities to transcribe real-time audio. The final code for the project is available on Github. To get started, we need to generate a Deepgram API key and install necessary dependencies such as Quart, Deepgram Python SDK, and dotenv library. We then create a basic Quart application with an HTML page and set up WebSocket connections between the server and browser, as well as between the server and Deepgram for real-time audio transcription. The final result is a live voice-to-text transcription project that can be run on localhost.

The team behind Stëmm developed a Chrome extension that leverages the Deepgram API and Chrome API to enable voice control for Google Chrome users with motor disabilities. The project, which was created during Hack Cambridge, allows hands-free browser control through voice commands like "chrome, open tab," "chrome, search for recipes," and "chrome, add bookmark." While the team faced challenges in meeting Google's strict security provisions for extensions, they successfully implemented the extension with the help of Deepgram's Speech Recognition API. The project is still being developed, and contributions are welcome on GitHub.

Technical writing is an essential skill for developers to communicate complex concepts effectively. It involves creating written and technical documentation for specific audiences. To get started, consider the audience's needs, expectations, and knowledge level. Focus on clarity, relevance, and providing practical examples. Break down technical writing into three categories: informational, instructional, and persuasive. Ensure your writing has a clear beginning, middle, and end with supporting ideas and examples. Proofreading is crucial to ensure accuracy and quality. Technical writing can be engaging if made interesting by the writer.

AutoBubble is a Chrome extension developed by first-year Computer Science students at the University of Cambridge during Hack Cambridge, a 24-hour student hackathon. The project aims to improve the display of captions for online videos using facial recognition and Deepgram's Speech Recognition API. AutoBubble places captions next to a speaker's face in a YouTube video, allowing for better understanding of the depth of expression. The team utilized documentation, tutorials, and example projects from Deepgram to build their project, which was highly praised at Hack Cambridge.

In this tutorial, the author demonstrates how to create a live voice-to-text transcription application using Python and Django frameworks along with Deepgram's AI speech recognition. The project involves setting up a virtual environment, installing necessary dependencies like Django, Deepgram SDK, dotenv library, and Django Channels. Then, the author guides through creating a Django project and app, integrating Django Channels, adding Deepgram API key, getting mic data from the browser, establishing WebSocket connections between server-browser and server-Deepgram, and finally running the application to get real-time transcriptions. The code for this project is available on Github with instructions on how to run it.

In this tutorial, we learn how to create a live voice-to-text transcription application using Python and Flask 2.0 with the help of Deepgram's AI speech recognition technology. The project involves setting up a virtual environment, installing necessary dependencies like Flask, Deepgram SDK, dotenv library, and aiohttp-wsgi. We then create a basic Flask application that renders an HTML page for live transcription. Next, we add the Deepgram API key to our project by creating a .env file and storing it in the GitHub repository's .gitignore file. The application uses JavaScript to access microphone data from the browser and establishes WebSocket connections between the Python server holding the Flask application and the client (browser). We also create a WebSocket connection between our central Flask server and Deepgram for real-time transcription. Finally, we run the application and start getting real-time transcriptions on the localhost. The complete code for this project is available on Github with instructions to run it.

In this tutorial, we learn how to create a live voice-to-text transcription application using Python and FastAPI. We use Deepgram's AI speech recognition for real-time audio transcription. The project involves setting up a virtual environment, installing necessary dependencies like FastAPI and Deepgram SDK, creating a FastAPI application, adding the Deepgram API key, getting mic data from the browser, establishing WebSocket connections between server and client, and finally processing audio to get live transcriptions. The final code for this project is available on Github.