April 2024 Summaries

This week's update highlights a new integration with Make.com, allowing users to automate audio data processing using AssemblyAI's Speech AI. The update also includes enhancements to streaming speech-to-text features such as session duration tracking and the option to disable partial transcripts. Additionally, fresh blog posts cover topics like redacting PII in audio with Make and AssemblyAI, transcribing audio files with Go and Python using Universal-1, and building an AI lecture assistant with Python. The update also includes a tutorial on creating video sections automatically with Python.

The text describes a tutorial on using Make (formerly Integromat), an automation tool, to create a scenario that watches a Google Drive folder for new audio files and then creates both a transcript and an audio file in which PII is redacted. The AssemblyAI app for Make allows users to process audio data by transcribing it with speech recognition models, analyzing it with audio intelligence models, and building generative features on top of it with LLMs. The tutorial guides users through setting up the necessary modules and connections within Make, including Google Drive Watch Files in a Folder as the trigger, Google Drive Download a File module, AssemblyAI Upload a File module, AssemblyAI Transcribe an Audio File module, AssemblyAI Get Redacted Audio of a Transcript module, and Google Drive Create a File from Text module. The tutorial concludes by encouraging users to explore additional features offered by the AssemblyAI app for Make.

Make is a workflow automation tool that enables users to integrate various services without coding knowledge. In collaboration with Make, AssemblyAI has made its advanced Speech AI available for no-code and low-code builders. The AssemblyAI app for Make allows users to process audio data using speech recognition models, analyze it with audio intelligence models, and generate features on top of it with LLMs. Users can supply audio to the AssemblyAI app and connect the output to other services in their Make scenarios. Each module in the AssemblyAI app has intuitive input parameters and detailed output parameters for easy integration. An example scenario includes transcribing, redacting PII, generating summaries using LeMUR, and uploading outputs back to Google Drive when a new audio file is added. To learn more about building with AssemblyAI and Make, visit the documentation or follow the tutorial on redacting PII using Make.

This week's update introduces an integration with LangChain Go, allowing users to leverage large language models (LLMs) for audio data analysis using AssemblyAI's speech-to-text models. The Universal-1 model, a multilingual Speech-to-Text model trained on 12.5 million hours of audio data, is now the default transcription model for all users due to its improved accuracy and performance capabilities. Additionally, new blog posts provide tutorials on transcribing audio files with Universal-1 using Go, Python, and Node.js.

This week's update introduces new features for streaming speech-to-text, including session duration tracking and the ability to disable partial transcripts. AssemblyAI has also achieved PCI DSS certification and GDPR compliance for secure data handling and protection of EU user data. Additionally, tutorials on transcribing audio files with Universal-1 in Ruby, Go, and Python are available, along with a guide on how to make video sections automatically with Python.

AssemblyAI has introduced Universal-1, a state-of-the-art speech recognition model that achieves near-human accuracy in transcribing audio files. Trained on millions of hours of data, it supports English, Spanish, French, and German languages with more to come. The company also launched Best and Nano models for applications where accuracy is paramount or cost-effectiveness is required, respectively. AssemblyAI's Ruby SDK allows developers to transcribe audio files using these models in their applications. Additionally, the platform offers features like entity detection, content moderation, PII redaction, and LeMUR for applying LLMs to audio data.

Transcription services are crucial in various fields such as legal, medical, and media. Traditional transcription methods often fail to meet the growing demands for speed, accuracy, and cost-efficiency. Advanced Speech AI technology uses artificial intelligence, machine learning, and natural language processing to deliver human-level accuracy that can understand multiple languages. This technology has several benefits including higher accuracy rates, faster turnaround times, scalability, cost savings, expanded market reach, customization and learning capabilities, security and privacy features, real-time transcription options, and improved accessibility for diverse audiences. Real-world examples showcase how companies like Screenloop, Aloware, and YouTube Transcripts have successfully integrated Speech AI technology into their services to enhance customer experiences, streamline workflows, and improve communication barriers.

AssemblyAI has introduced Universal-1, a state-of-the-art speech recognition model that achieves near-human accuracy in speech-to-text transcription. Trained on millions of hours of audio data, it supports English, Spanish, French, and German, with more languages to be added soon. The company also launched two new classes of models: Best for applications where accuracy is paramount, and Nano as a cost-effective alternative supporting 99 different languages. Users can transcribe audio files in their Java applications using the AssemblyAI Java SDK.

This week, AssemblyAI launched Universal-1, their most powerful and accurate Speech-to-Text model to date. Trained on 12.5 million hours of multilingual audio data, it offers improved speaker count estimation, word timestamp estimation, fewer hallucinations, and more accuracy compared to competitors' speech-to-text APIs. The new model can transcribe multiple languages within a single audio file and processes an hour of audio in just 38 seconds. Additionally, AssemblyAI released tutorials on using Universal-1 with Go and Python for transcription tasks and building AI voice bots from scratch.

LangChain is a framework that helps develop applications using Large Language Models (LLMs). It provides common building blocks for integrating with LLMs. However, LLMs only work with textual data and do not understand audio data. Recently, an AssemblyAI document loader was added to LangChain Go, enabling the integration of AssemblyAI's speech-to-text models. This feature is also available for both Python and JavaScript versions of LangChain. An example demonstrates how to transcribe an audio file using AssemblyAI and generate a response to a question using OpenAI. LeMUR is another framework that leverages LLMs to understand speech, especially when working with audio data.

AssemblyAI has introduced Universal-1, a new speech model that sets a standard for automated speech recognition (ASR) accuracy. The model is designed to transcribe accented speech, background noise, and difficult phrases with near-human accuracy. It can be accessed through the same web API as previous ASR models. Alongside this release, two new pricing tiers have been introduced: Best and Nano. A tutorial demonstrates how to use Python applications to transcribe audio or video files using Universal-1's Best and Nano tiers with AssemblyAI's Speech-to-Text API. The tutorial also explains how to switch between the Best and Nano tiers by adjusting the TranscriptionConfig parameters.

AssemblyAI has introduced a new speech model called Universal-1, which boasts near-human accuracy in speech-to-text conversion even with accented speech and background noise. The model is also faster than its predecessor, Conformer-2, and supports English, Spanish, French, and German languages. Alongside this release, two new pricing tiers - Best and Nano - have been introduced. The Best tier offers high accuracy for applications where precision is crucial, while the Nano tier provides cost-effective transcription services in 99 different languages. Users can transcribe audio files using Universal-1 and Nano through AssemblyAI's Go SDK by setting up an authenticated client and configuring speech model parameters.

Contact center speech analytics leverages artificial intelligence to transform raw voice data into actionable insights, enabling businesses to analyze customer interactions at scale and gather real-time behavior data. AI-powered speech analytics tools can help identify customer complaints or issues, pinpoint upsell and cross-sell opportunities, guide training programs for employees, ensure compliance and quality assurance, reveal trends in customer behavior, preferences, and market demand, and understand customer sentiment and emotional tone. These insights can significantly boost a contact center's results by improving customer understanding, optimizing agent performance, enhancing operational efficiency, identifying sales and upsell opportunities, reducing manual post-call work, maintaining compliance, mitigating risks, protecting employees, and discovering new customer insights.

This weekly update highlights new product features, tutorials, and community resources related to LeMUR Cookbooks for building audio LLM applications. It also introduces a free Python course on creating an audio AI app using AssemblyAI's speech-to-text capabilities and LeMUR's LLM functionalities. Additionally, the update covers blog posts on extracting insights from phone calls with LLMs in Python, AI trends in 2024 focusing on Graph Neural Networks, and how to do Speech-To-Text with Go. Lastly, it mentions popular YouTube tutorials on building a RAG application for multi-speaker audio data and coding an AI voice bot from scratch using Python.

AssemblyAI has launched Universal-1, a multilingual speech-to-text model trained on 12.5 million hours of audio data in four major languages - English, Spanish, French, and German. The model boasts best-in-class accuracy, reduced word error rate, improved timestamp estimation, and is designed to handle various conditions such as background noise, accented speech, natural conversations, and language changes. Universal-1 has shown significant improvements over previous models in terms of accuracy, hallucination reduction, and precise timestamp estimation. The model's capabilities are expected to power the next generation of AI products and features built with voice data.

AssemblyAI has introduced a new speech model called Universal-1, which boasts near-human accuracy in speech-to-text transcription even with accented speech and background noise. It is significantly faster than the previous model, Conformer-2, and supports English, Spanish, French, and German, with more languages to be added soon. Alongside this new model, two new pricing tiers have been introduced: Best for applications where accuracy is paramount, and Nano as a cost-effective option supporting 99 different languages. The AssemblyAI Go SDK can be used to transcribe audio files in Go applications using Universal-1 and Nano.

Universal-1 is a highly advanced speech recognition model developed by AssemblyAI. It has been trained on an extensive dataset of over 12.5 million hours of multilingual audio data, making it the most powerful model in its class. The model boasts best-in-class accuracy for speech-to-text conversion across four major languages: English, Spanish, French, and German. This cutting-edge technology is set to revolutionize industries that rely heavily on accurate speech recognition capabilities.

Universal-1 is a highly advanced speech recognition model developed by AssemblyAI. It has been trained on an extensive dataset of over 12.5 million hours of multilingual audio data, which enables it to achieve top-tier accuracy in speech-to-text conversion across four major languages: English, Spanish, French, and German. This cutting-edge model is a significant advancement in the field of speech recognition technology.