Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
TL;DR: Native Microsoft Teams transcription via the Graph API delivers transcripts only after a meeting ends, provides utterance-level (not word-level) timestamps, and degrades sharply on accented or multilingual speech. For any product that routes Teams audio to downstream AI systems, the more reliable architectural pattern is capturing raw audio via a custom WebRTC bot and routing it to a managed STT engine, one that delivers word-level timestamps, accurate multilingual handling, and predictable per-hour costs, none of which the native Graph API provides. Solaria-1 covers real-time streaming and broad language support. Solaria-3 is optimised for European business audio.
Podcast transcription at scale: an API workflow for media platforms
TL;DR: Podcast audio is invisible to search without accurate, word-level transcripts, and transcription quality sets the ceiling for everything downstream, from content discovery to AI-generated show notes. A production-grade async pipeline (decoupled webhook ingestion, pyannoteAI-powered diarization, word-level timestamps) is what separates a searchable audio library from a title-and-description catalog. At 10,000 hours monthly, a managed API costs $2,000–$6,100 depending on plan.
HIPAA-ready meeting assistants for healthcare and therapy sessions
TL;DR: Building a HIPAA-ready meeting assistant requires more than a generic transcription wrapper. Any API that processes Protected Health Information on your behalf must sign a Business Associate Agreement (BAA) before PHI flows to it, and transcription accuracy matters more than most teams expect: word error rate can more than double in noisy, multi-speaker clinical environments compared to controlled recordings, meaning errors compound into every SOAP note and EHR entry downstream. This guide covers the BAA requirements, encryption controls, and unit economics product teams need to evaluate before committing to an audio infrastructure provider for clinical or therapy use cases.
How Selectra is automating quality monitoring of sales calls with speech-to-text AI
Published on Sep 9, 2024
In the past few years, the democratization of speech recognition and large language models has created new opportunities for voice-first platforms to automate critical workflows. Customer support is one of the most promising and vibrant areas for these innovations.
Selectra, a leading utility comparison and sales company, operates an extensive in-house fleet of on-call sales representatives and is actively embracing the new paradigm to improve agent productivity and customer experience. The company uses Gladia AI's speech-to-text engine to improve quality monitoring and extract in-depth insights from every call, continuously refining its customer experience. Here comes their story.
About Selectra
Selectra was founded in 2007 in France with the aim of helping consumers choose among the many electricity providers emerging on the market.
Since then, the company pursued rapid growth, reaching more than 200 million unique visitors a year to its websites. Now active in 17 countries, it has expanded across verticals – insurance, telecom, banking and others – and now caters to both individual consumers and companies through educational content and dedicated call center support.
Challenge
The company receives thousands of inbound calls daily from customers who want to learn more about the different providers. During the call, the call agent’s goal is to suggest the best offer for the customer. In France alone, the company has over 200 sales agents processing calls daily.
In this context, quality monitoring is key to consistently improving customer experience. The company’s dedicated quality assurance team is responsible for reviewing the quality of the call based on pre-set criteria.
Historically, the team would process the calls one by one, which involved downloading the call and reviewing each transcript in search of specific words and phrases to ensure compliance with internal protocols.
The process was time-consuming, but with the emergence of Large Language Models (LLMs), Selectra could automate it in part by feeding the call transcripts to the model and letting the latter determine the degree of compliance and extract necessary insights.
This substantially accelerates the work of the human agents, as all they need to do in this case is validate the model’s findings using Selectra’s internal platform, built specifically for the new workflow.
In parallel, the company began to leverage speech-to-text technology to acquire more in-depth insight into customer needs and sentiments and analyze sales reps’ performance in more detail.
Objectives
To make all of the above possible, Selectra looked for a speech-to-text API to provide a solid foundation for the internal quality assurance and analytics platform, combining the following:
Top-quality async transcription, free of hallucinations, to ensure the best possible input for LLMs.
Impeccable accuracy in French, with the prospects of expanding to other languages.
Transcript displayable in sentences (as opposed to utterances), with correct punctuation and named entities
Solution
Enter Gladia! With Gladia, Selectra was able to implement the following product features and improvements:
Enhanced search experience, enabling agents to easily locate a specific segment of a call thanks to error-free capture.
Automated quality monitoring, where transcription is used to record the call and match the output to the original script to assess the degree of compliance, with human agents simply having to validate the final assessment.
Extracting topics and sentiments, to determine how customers feel on the call and identify recurring issues.
Preview of Selectra's internal platform for quality monitoring
Impact
After deciding to add audio transcription to the new quality monitoring project, Selectra started testing the API immediately. By working with the Gladia team to iterate and scale up, they saw a noticeable impact on the efficiency of the quality assurance workflow:
Today, the Selectra team is already considering how they can leverage Gladia’s transcription further with the audio intelligence add-ons (like chapterization, NER and others), as well as our product's real-time capabilities.
As speech recognition and LLMs become more advanced, Selectra is convinced that metadata from calls will be used in multiple new ways moving forward, enabling them to extract truly granular insights from every customer interaction. We thank Selectra for their trust and looking forward to pushing the boundaries of what speech recognition together.
About Gladia
Gladia provides a speech-to-text and audio intelligence API for building virtual meeting and note-taking apps, call center platforms, and media products, providing transcription, translation, and insights powered by best-in-class ASR, LLMs and GenAI models.
Having read this case study, do you feel like Gladia could be the right fit for your business too?
.png)
