Gnani AI releases Prisma v3.0 speech-to-text for code-mixed Indian phone calls
Bengaluru-based Gnani AI launched Prisma v3.0, one speech-recognition model for nine Indian languages plus Indian English, trained on 8 kHz telephony audio and heavily code-mixed speech. It offers verbatim, code-mix and transliteration modes through REST, streaming and batch APIs. Gnani's own Vistaar evaluation puts its word error rate below ElevenLabs Scribe v2.
Why it matters
Teams building voice bots for South Asian call centres get a telephony-tuned, code-mix-aware ASR option, though the benchmarks are self-reported.
Source: Read the full story on Gnani AI