August 2022 Summaries

The article discusses how to monitor media mentions in podcasts using Python and AI speech-to-text technology. It presents two methods for this purpose, diarization with AI speech recognition and search and entity detection with Python, Deepgram's AI speech-to-text Search feature, and SpaCy. The latter method is considered more effective as it involves searching for terms or phrases in the podcast transcript that indicate an ad segment, identifying the organizations mentioned in the ad to determine the company sponsoring that segment, and using entity detection with SpaCy to tag recognized company entities as ORG. A Python code breakdown is provided to illustrate how these methods can be implemented.

Text cleaning is an essential component of natural language processing that helps prepare training data for automatic speech recognition (ASR) systems. It involves transforming raw data into a "cleaner" version, closer to the actual phonetics of what was said. This process is language-dependent and requires a multi-step processing pipeline to ensure accurate transcriptions. In Turkish text cleaning, challenges include handling the apostrophe, consonant assimilation, vowel harmony rules, and processing currencies and numbers. Text cleaning is crucial for ASR training as it helps improve the accuracy of transcriptions by ensuring a good match between phonetics and phonetic transcription.

In this tutorial, learn how to transcribe phone calls using Twilio Functions and Deepgram. First, set up a Twilio account with a phone number that has Voice capabilities. Next, create a new service in the Twilio Console and add @deepgram/sdk as a dependency. Add your Deepgram API Key as an environment variable. Create two functions: /inbound and /transcribe. The /inbound function will prompt callers to speak after the beep and record their message. When the call ends, data is sent to /transcribe. In the /transcribe function, use Deepgram's Node.js SDK to generate a transcription of the recording. Then, send the transcript to the caller via SMS using Twilio Client. Finally, deploy both functions and test by calling your Twilio number, speaking after the beep, and hanging up. You should receive an SMS message with the transcribed text shortly afterward.

Michael Jolley shares his journey of learning Python by building a self-described "useless" but educational podcast search engine utilizing Flask for the back-end and Jinja2 for templating. Inspired by Simone Giertz's TED Talk on the value of creating non-functional projects, Jolley explores Python's capabilities in handling HTTP requests, file access, and audio processing. He utilizes Deepgram's Python SDK to transcribe podcast episodes and Flask to respond to search queries, creating a basic search engine that finds and displays phrases within a podcast transcript. Despite the intentionally simplistic and unattractive user interface, the project effectively demonstrates key programming concepts such as text normalization and API interaction, culminating in an audio player feature that allows users to play specific segments of the podcast around their search terms.

This tutorial demonstrates how to transcribe podcast feeds using terminal commands. It requires a Deepgram API Key and the installation of jq and yq for XML manipulation. The process involves extracting specific items from an RSS feed, displaying specific properties, looping through objects, and generating transcripts using cURL requests. The final result is text files containing transcriptions of podcast episodes.

Content moderation plays a crucial role in keeping offensive language out of various platforms such as video games, forums, ad campaigns, etc. Speech-to-text technology supports content moderation by converting speech to text, enabling the same processes used for written content to be applied to spoken content. This helps in identifying problematic content like sexually charged messages, hate speech, radicalization, bullying, scams, grooming, and more. Automatic moderation using AI-powered automatic speech recognition solutions can help in reducing human intervention and making the process efficient. Content moderation is used across various industries including gaming, forums and social media, advertising, ecommerce, health care, and finance to ensure compliance with site guidelines and relevant laws.

This tutorial demonstrates how to transcribe podcasts using Python and the Deepgram API. The process involves installing necessary dependencies, setting up a virtual environment, and creating an empty script file. Users can transcribe hosted files, local files, or podcast RSS feeds by adjusting the source property in the code. Additionally, speaker detection and paragraph formatting can be enabled to improve transcript quality. Finally, the generated transcript can be saved to a text file for further use.

In this blog post, the author discusses a Python project for AI Machine Learning Topic Detection using podcast audio files. The main steps involved are transcribing speech-to-text with Deepgram's API and then applying the TF-IDF (Term Frequency - Inverse Document Frequency) topic detection algorithm to identify key topics in the podcast episode. The author provides a detailed walkthrough of the Python code, including functions for removing stop words, vectorizing cleaned documents, and performing K-Means clustering to create 10 clusters of topics. The final results are written to a file called "results.txt" for further analysis.

Real-time streaming transcription involves transcribing live audio into text, with applications such as live captioning for the hearing impaired or enabling machines to understand human speech. The process is similar to pre-recorded transcription but requires different input and output configurations. Key metrics for evaluating real-time streaming transcriptions include Word Error Rate (WER), Word Recall Rate/Word Recognition Rate (WRR), and ASR latency. Deepgram offers several benefits over other ASR providers, including high accuracy, low ASR latency, support for all languages and use-case models, and the choice of cloud or on-premise deployment.

This tutorial demonstrates how to transcribe YouTube videos using the terminal with the help of Deepgram's Speech Recognition API. The process involves downloading audio from a YouTube video, transcribing it with Deepgram's API, and then extracting the transcript text. To achieve this, you will need to install youtube-dl, ffmpeg, jq, and obtain a Deepgram API Key. The steps are similar to transcribing YouTube videos using Node.js but performed entirely on the terminal.

This tutorial demonstrates how to use Python and the Deepgram API speech-to-text audio transcription to play a piano with your voice. The project requires macOS, but it is also possible with Windows or Linux machines. We'll use Python 3.10 and other tools like FluidSynth and Deepgram Python SDK speech-to-text audio transcription. The code allows users to record their voice and transcribe the spoken numbers into notes on a virtual piano, playing the first few phrases of Lady Gaga’s Bad Romance as an example.

The Complete Guide to Punctuation & Capitalization in Speech-to-Text discusses how automatic speech recognition (ASR) can be tricky for punctuation and capitalization, especially across different languages. It explains the purpose of punctuation and its role in making a text more understandable by providing some sense of intonation and pacing that would occur if a sentence was spoken out loud. Capitalization is also discussed as the process of making a letter capital or upper case, with proper nouns being typically capitalized in English. The importance of punctuation and capitalization for speech recognition is highlighted, as they make texts clearer and easier to read by humans. Two methods of adding punctuation to ASR transcripts are presented: a separate punctuation and capitalization model that runs after the text has been generated, and an end-to-end deep learning system that generates both text and punctuation simultaneously, using acoustic information for better accuracy.

Speaker diarization is a process that separates individual speakers in an audio stream, allowing each speaker's utterances to be separated and labeled with their unique audio characteristics. This feature can also be called speaker labels or speaker change detection. It is used to increase transcript readability and better understand what a conversation is about. Common use cases for speaker diarization include audio/video management, compliance, conversational AI, education, health, law enforcement, recruiting, sales enablement, and speaker analysis. The main metric used for speaker diarization in the business world is the accuracy of identifying individual speakers or "who spoke what." Deepgram's speaker diarization has several benefits, including no need to specify the number of speakers in the audio, no cap on the number of speakers, support for any language Deepgram transcribes, and support for both pre-recorded or real-time streaming audio.

In this tutorial, the author demonstrates how to use cURL and jq commands to extract transcripts from audio files using Deepgram API. The process involves making a POST request with appropriate headers and data, then using jq to filter out the desired information (transcript in this case). Additionally, the author explains how to save the output to a file and create bash scripts for processing multiple audio files. Finally, they showcase some advanced features of jq that can be used to format the extracted transcripts.

In this text, the author reminisces about the Cloud to Butt Chrome Extension and shares a nostalgic experience using the Deepgram Find and Replace feature to replace instances of "the cloud" with "my butt" in an audio file. The author demonstrates how to use cURL and jq command-line utilities to manipulate JSON data and display the transcript after replacing words in the audio file. They also encourage readers to provide feedback on their post through GitHub discussions.

Closed captioning is an essential feature that helps people who are deaf or hard-of-hearing to follow along with video content, and it's increasingly being used by others as well. Speech-to-text technology and automatic speech recognition (ASR) play a crucial role in enabling closed captioning for various types of media, including live events, television broadcasts, educational videos, podcasts, and more. By using AI-powered solutions like Deepgram, companies can provide accurate and real-time captions that improve accessibility and user experience across different domains.

In this tutorial, learn how to build a Pipedream workflow that automatically transcribes Google Drive files using the Deepgram Node.js SDK and sends the transcriptions via email. To begin, create an empty Pipedream workflow and set up triggers for new files added to a specific Google Drive folder. Next, download the file to Pipedream's temporary storage and transcribe it with Deepgram. Finally, send the transcription via email and clean up the storage by deleting the file from the tmp directory. This tutorial provides step-by-step instructions for setting up each part of the workflow and testing its functionality.

Contact Center as a Service (CCaaS) platforms leverage automatic speech recognition (ASR) solutions to automate and streamline customer inquiries processing. ASR technology converts speech into written text, enabling computers to understand spoken language. CCaaS companies use these solutions for call routing, customer support, and transcribing call recordings. The benefits of using CCaaS platforms include lower IT costs, improved customer satisfaction, advanced call tracking and analytics, and easier call routing. Deep learning-based ASR tools are particularly effective in contact center applications due to their ability to adapt to industry jargon and poor call quality.

Meeting analysis platforms utilize automatic speech recognition (ASR) solutions to transcribe meeting minutes into text, saving time and improving decision-making and communication within organizations. These platforms offer features such as separation of speakers, topic detection, summarization, and search capabilities. The benefits include action item lists, record decisions, time savings, improved decision-making, better communication, easier note-taking, and taking attendance. Key factors to consider when choosing a meeting analysis platform are ease of use, pricing, and accuracy. Automatic speech recognition (ASR) is crucial for these platforms as it enables the automatic transcription of conversation contents.

This article provides a comprehensive guide on using the Python library DeepSpeech for speech-to-text conversion. It begins with an overview of DeepSpeech, which is an open-source software inspired by Baidu's 2014 paper and currently maintained by Mozilla. The author then delves into how to set up DeepSpeech locally for transcription purposes. The setup involves installing necessary libraries such as deepspeech, numpy, and webrtcvad using pip. Three files are required for the asynchronous transcription: one each for handling WAV data, transcribing speech to text on a WAV file, and using these two in the command line. The first file, wav_handler.py, is responsible for reading and writing audio data from/to a WAV file, creating frames of audio data, and detecting voice-activated frames for speech recognition with DeepSpeech. The second file, wav_transcriber.py, transcribes speech to text for a WAV file using DeepSpeech models. It also includes functions to load the pre-trained model into memory and generate VAD segments. The final part of the guide focuses on creating a command line interface (CLI) for real-time and asynchronous speech recognition with DeepSpeech. The CLI allows users to pass in options to choose whether they want to do real-time speech recognition or run speech recognition on an existing WAV audio file. The article concludes by summarizing the steps involved in setting up DeepSpeech for local transcription and using it through a command line interface.