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.
MiFID II and FCA call recording: compliance for voice transcription in finance
TL;DR: Financial firms operating under MiFID II and FCA jurisdictions must maintain searchable, high-accuracy records of all client-related voice communications, including remote and mobile calls under FCA Market Watch 66. Engineering and product teams building for these obligations typically implement dedicated cloud infrastructure with defined data residency, controls that prevent customer audio from being used to retrain models on Growth and Enterprise plans, and transcription accurate enough that the resulting records hold up under regulatory review. Transcription and speaker attribution errors are not product quality issues in this context. They are audit trail failures, and regulators treat them as such.
Why speech-to-text accuracy matters upstream of your LLM
TL;DR: Downstream LLM performance is ceiling-bounded by upstream transcription accuracy. A transcript with a meaningful error rate doesn't produce proportionally degraded summaries or CRM entries. It produces outputs where hallucinated names, inverted logic, and misattributed speaker turns compound silently into every downstream system that reads them. Prompt engineering cannot recover information the STT layer never captured. Gravite cut call quality review time by 93%, from 15 minutes to 1 minute per call, once transcript accuracy was high enough to trust the output without manual verification.
TL;DR: Telephony audio is constrained to 8kHz narrowband frequencies, stripping away the high-frequency spectral energy that generic 16kHz STT models require. Standard upsampling cannot recover phonemes that were never captured at the source, leading to transcription errors that compound silently into downstream NLU failures. Solaria-3 ranks #1 on Switchboard, the most challenging conversational telephony dataset, ahead of AssemblyAI, ElevenLabs, Deepgram, Mistral, and Speechmatics, ensuring your downstream LLM pipelines receive clean, structured data from real-world noisy call audio.
Mastering AI transcription for social media captions: Mojo's success story with Gladia
Published on Dec 17, 2023
From Reels to ads to YouTube shorts, video content consumed in vertical bite-size format on social media is becoming among the primary ways we interact with the world for both leisure and business.
It has been estimated that in 2023 people globally have watched, on average, 17 hours of online video per week (that is, at least one hour daily) — and there’s no reason to expect the trend to subside in the future.
In this context, high-quality automatic captions for videos are becoming a must-have feature for video marketing, given that most consumers expect to be able to consume video on the go without sound. Captions are likewise incredibly important to ensure the accessibility of one’s social applications to all types of users.
Mojo, which specializes in social media content editing, is among our clients relying on speech-to-text to generate high-quality automatic captions and other voice-based editing features.
About Mojo
Mojo is a social video and content app, designed to make high-quality social media content creation easy. With its one-stop-shop app, Mojo users can access hundreds of templates, text effects, and high-quality animations to create stunning social content in a matter of minutes.
Founded in France in 2018, the company targets primarily small businesses in search of easy-to-use tools to showcase and promote their brand across social media channels with Reels, stories, TikTok, posts and more. In the last four years, the app has gained over 46M users, including several hundred thousand paying subscribers.
Challenge
Mojo’s team knew that delivering on its mission meant that they had to make video content editing easy, intuitive, and feature-rich. To stand out in a globalized, ultra-competitive market, the company focused on building a library of hundreds of animated templates and delivering advanced editing tools and effects, such as video trimming and background removal.
The integration of audio transcription in particular into Mojo's app came in response to growing user demand for auto-captions, aligning with the company’s ongoing shift from template-based to tool-based app usage.
Having tried several alternative speech-to-text APIs, Mojo’s team turned to Gladia in search of a better solution.
The main issue encountered with the incumbent provider was that it did not meet the quality standard in word timings, aka word-level timestamps, which were especially critical for the right timing of visual effects.
In addition, the language support of the previous provider was not always reliable to the detriment of client satisfaction and retention.
Objectives
To deploy a high-quality, scalable transcription and audio intelligence API to power the Mojo app, with the following specifications:
Accurate and fast batch transcription for auto-captions at a scalable cost.
Language support (especially in European languages), to serve the app’s global client base, including the ability to automatically detect not only one’s language but also dialects and accents.
Top-level precision for word-level timestamps, with the start and end times of each word, detected perfectly, being an essential pre-requisite for video and captions editing.
Solution
Enter Gladia. With Gladia, the Mojo team was able to enhance in-app user experience with the following features:
Auto-captions, supported in multiple languages, enhanced with a style selection tool.
Remove pauses, where time-stamped transcripts are used to automatically detect and remove silences from a video.
Mojo UI for auto captions
Auto captions as seen inside the Mojo app
To enable these features, Gladia relies on an advanced hybrid model architecture with generative AI components. Not only does our API detect words accurately based on acoustic properties of speech, but it also fills the gaps in the transcript based on a contextualized understanding of language, while deploying a range of techniques to eliminate hallucinations and accurately detect a variety of accents even in complex environments.
Impact
By working with the Gladia team to iterate and scale up the volume to thousands of hours transcribed monthly, Mojo saw a noticeable impact on the usage of the auto captions feature, with happy users expressing their satisfaction with the new spot-on quality of the feature.
The Mojo team is continuing to explore the possibilities that transcription brings to the Mojo products and services, and is now considering how they will leverage it in the future with upcoming features like keyword detection.
We're thrilled to be part of this amazing journey with them, and thank Mojo for putting their trust in us! As we move forward, we're excited to team up with more clients, tackle new challenges, and make speech AI more accessible to media companies worldwide.
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?
.gif)
