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.
Opening up new markets for a sales meeting and CRM enrichment platform: Spoke's success story with Gladia
Published on Feb 27, 2024
In the past, sales teams around the world were presented with a twofold challenge. In addition to showcasing their products in the best light to prospects, they needed to take detailed notes during the call and fill their CRM software manually afterward.
The whole process was challenging, time-consuming, and inefficient, taking away the valuable time of the customer-facing staff from tasks they do best and deliver the most impact in.
Luckily, that is changing with the help of AI-powered virtual meeting platforms optimized for sales teams. Spoke is among the leading French platforms contributing to the CRM automation trend in the field with the help of Gladia’s multilingual audio transcription API.
About Spoke
Founded in Paris in 2020, Spoke is on a mission "to transform any salesperson into the best salesperson ever" with the help of its versatile meetings AI assistant.
Spoke’s platform is distinguished by a highly customizable interface, allowing users to set a wide range of custom parameters and ask the GTP interface questions, both before and after the meeting, to optimize the output and derive in-depth insights respectively.
The company’s primary target is customer-facing salespeople — the ultimate end users of Spoke — with the management staff being equally incentivized to engage with the platform to review the ideas and for quality assurance purposes.
To get started, Spoke offers out-of-the-box templates, based on the most common sales frameworks like BANT, which can be customized extensively for more granular company-specific knowledge. It’s this “granular extraction” that the CEO of Spoke, Lazare Rossillon, considers the product’s most powerful point of difference today — and we couldn’t agree more.
Challenge
In order to enrich CRM, perfect accuracy is indispensable — after all, CRMs are the single source of truth for companies that have a direct impact on their future revenue. The quality of the CRM is then as good as the quality of its underlying transcription engine, with errors like wrong speech attribution or misspelled personal details risking to contribute to a drop in customer satisfaction and missed deals.
Lazare and the team needed a highly accurate and reliable transcription in French to extract data from calls and automatically fill up CRM databases like Salesforce or HubSpot.
At first, the team tried working with both the Big Tech speech-to-text solutions and specialized contenders but found them predominantly US-centric. With a predominantly French-speaking user base, the company needed much better performance and quality in English and French.
Objectives
To deploy a high-quality speech-to-text API and audio intelligence add-ons for all Spoke customers to transcribe and analyze online sales meetings and user interviews.
Specifications required:
Error-free batch transcription, powered by a model that removes virtually all hallucinations from transcripts to enable successful CRM enrichment.
Multilingual support, especially in French and other European languages;
Custom vocabulary, allowing to introduce specific company names and terminology to ensure they are transcribed accurately in every conversation.
Solution
With Gladia, Spoke was able to ensure a smooth user experience across a range of its core functionalities:
Searchable transcripts of each meeting, generated in near real-time in the form of an editable Q&A-styled report as the meeting is ongoing;
Interactive video playback, with key highlights of the meeting linked to the corresponding transcript sections;
Highly customizable dashboard, with the ability to pick a template for organizing the transcribed notes and set a custom vocabulary for them.
Impact
Setting up Gladia took the Spoke team less than a few days of work, thanks to the API’s comprehensive documentation and the helpful team in charge of customer support.
The results did not take long to wait, with Spoke users noticing a huge impact of the new transcription provider on their services.
Over 27,000 meeting hours transcribed by Spoke users, with an average of over 1,500 hours a week;
Number of errors have reduced drastically, as testified by Spoke users;
New European markets opening up thanks to Gladia’s best-in-class multilingual support.
The team at Spoke is now considering how they will leverage our API in the future. Spoke’s ultimate aspiration is to enable a two-way synch of information on their platform, with advanced AI assistance features that will serve to augment sales teams around the world with actionable, real-time insights and tips, based on client history and preferences.
Expecting that speed is going to become increasingly important for CRM enrichment moving forward, Spoke is confident that Gladia’s industry-leading latency in both asynchronous and live transcription can be counted on to deliver on this vision.
Update: Our friends at Spoke have just built a great new tool, Meeting BAAS, for meeting recordings. Don't miss it!
Having read this case study, do you feel like Gladia can be the right fit for your business too?
