gensai.aiSpeech recognition, or speech-to-text (STT), transcribes speech into text. Models like Whisper have made it accurate, multilingual and runnable locally. It is the entry point of any voice agent: transcription quality conditions everything that follows.
In practice at Gensai
Gensai runs faster-whisper locally for its voice installations, such as the Oceans device, and AssemblyAI for Qualicontact's donation-call analysis.
Local or API: which to choose?
A local model, such as faster-whisper, suits systems without a network or sensitive data. A specialised API, such as AssemblyAI, adds advanced features (speaker separation, timestamps, high-volume processing) that are useful for analysing thousands of calls. The right choice depends on confidentiality, volume and expected latency.
Common mistake
Testing on recordings that are too clean. Background noise, accents, business vocabulary and proper names strongly degrade transcription. The model must be evaluated on real recordings, in real conditions of use.
Same theme
Definition written with AI assistance, checked by Sébastien Lecocq, co-founder of Gensai.
Going beyond the definition?
From concept to project: Gensai builds custom, sovereign, GDPR-compliant AI solutions.
Related services: Voice AI, Chatbots and answer engines