AI is finally learning to listen instead of just hear. A new speech-to-speech sentiment analysis startup aims to be the one to bring true emotional intelligence to the masses.
First-generation text-to-speech systems are giving way to second-generation, speech-to-speech AI designed to capture not just what people say, but how they say it. Established players like Qualtrix, Listen Labs, and Outset focus on enterprise clients with deep pockets.
But ReadingMinds.ai, a new voice-native sentiment analysis platform, is coming out of stealth with a different mission: deliver richer, real-time sentiment analysis to the SMB and midmarket crowd. It aims to capture not just words but how a customer says them — the tone, the pacing, and the pitch that reveal their true emotions.
“In the evolving landscape of AI-driven communication, preserving the human element (the tones, the emotions, the nuances) is critical — and that’s exactly what a unified speech model empowers you to do,” Stu Sjouwerman, CEO of ReadingMinds.ai, told TechRepublic.
The limits of old-school voice AI
First-generation voice AI systems were fairly effective. But as any user of Alexa or Siri can testify, they were far from perfect. Why? These systems take a linear, multi-step approach.
One model transcribes someone’s Speech To Text (STT), a large language model (LLM) analyzes the text, and the analysis results are converted to Text-To-Speech (TTS). Sjouwerman noted that this piecemeal pipeline of STT→LLM→TTS adds complexity and loses vital vocal nuances, such as emotional responses and other sentiment signals conveyed by the human voice. This first-gen approach loses that context when it flattens voice into text.
Instead of handing off between modules, ReadingMinds uses a neural model that listens and understands in a single pass. This preserves and leverages voice and emotional signals throughout processing. Further, it uses these inputs to dynamically adjust and respond to a user or customer based on intonation and rhythm.
“The result is faster, more natural understanding with shorter wait times, since the AI isn’t slowed down by having to go through multiple components,” said Sjouwerman. “By streamlining what used to be a multi-step pipeline into a single intelligent loop, we maintain the full richness of the spoken input — what was said and how it was said — to power more nuanced sentiment analysis.”
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
A more human experience
Sjouwerman’s company uses the Amazon Web Services Nova Sonic Model, which contains a massive amount of Alexa and other voice interactions. Add to that ReadingMinds’ secret sauce — how it arranges user prompts and how the AI agent conducting an interview is trained to respond and adjust in real time.
What you end up with is a system that detects emotions from acoustic cues. This enables the agent to mimic listening to the respondent in a way very similar to a human, and adjust its approach to better match the person being interviewed.
“The AI agent can sound sympathetic if it detects frustration or will stay calm if it senses the user is anxious,” said Sjouwerman. “Emma, your ReadingMinds AI interviewer, isn’t just transcribing and replying. It understands a customer’s mood and adjusts its answers on the fly, while also providing data-backed insights.”
By providing an experience that feels more human and supportive, companies gain real-time sentiment data and intent signals that inform better decisions — from tailoring a support call on the fly to identifying high-intent buyers.
An emotionally intelligent approach is instrumental in surveying an audience about a particular topic. The system is set up to instantly formulate questions specifically related to the area to be addressed. The AI agent interviews someone, adjusts its emotional state accordingly, and generates follow-up questions on the fly based on the answers. It also tabulates survey responses automatically.
ReadingMinds has released a beta version, which is currently available. Broad release is planned for the beginning of 2026. According to Sjouwerman, “the price will be a no-brainer” to make it accessible to mom-and-pop shops, SMBs, and the midmarket.
ReadingMinds.ai’s push for more human-sounding agents mirrors a broader shift toward voice-native tools, including Deepgram’s real-time voice intelligence for AWS Marketplace sellers, which converts raw speech into actionable insights.