DetectifAI, an Indian artificial‑intelligence startup established in 2025 by entrepreneur Tarini Padmanabhuni, announced the rollout of a live‑call detection system designed to combat the growing threat of AI‑generated voice deepfakes. The company describes its offering as a "voice‑trust layer" that works in real time to confirm the identity of the person speaking, analyse whether the speech is synthetic or a replayed recording, and raise an alert if the interaction appears suspicious. By embedding the technology directly into the call‑handling infrastructure, the system can intervene before a transaction is finalized or a sensitive conversation proceeds.
The core of DetectifAI's solution combines acoustic fingerprinting, neural‑network‑based synthetic‑speech detection and behavioural voice analytics. When a call is initiated, the platform captures the audio stream, extracts unique vocal characteristics and compares them against a verified voice profile stored by the client. Simultaneously, it runs a deep‑learning model trained on thousands of synthetic and replayed samples to spot anomalies such as unnatural intonation, frequency patterns or timing glitches that are typical of AI‑generated voices. If the model flags a high probability of manipulation, the system can either pause the call, request an alternative authentication factor or notify a human operator, thereby preventing fraud before it happens.
ALSO READ | India and ASEAN Nations Chart Scalable Agricultural Cooperation at Summit
DetectifAI is focusing its go‑to‑market strategy on sectors where voice verification is already a regulatory requirement or a common fraud vector. Early pilots with several Indian banks, KYC service providers and major telecom operators have demonstrated a reduction in fraudulent call‑based incidents, according to the startup's internal testing. The company says its technology can be integrated via APIs into existing IVR systems, mobile apps or contact‑center platforms, offering a scalable defence against increasingly sophisticated voice‑cloning attacks.
