Gnani AI has launched Timbre v3.0, a speech-generation model with inline tags that direct delivery rather than being read aloud. Writers can cue a whisper, laugh, sigh or pause while the model handles text in nine Indian languages and Indian-accented English. Number processing distinguishes amounts from account digits, with business calls and creative narration among the intended uses.
Gnani AI, a company developing speech technology, launched Timbre v3.0 on 14 October. The text-to-speech model takes short instructions embedded in a script and uses them to control delivery. A writer can indicate where the voice should whisper, laugh, sigh, slow down or pause. The model performs the instruction on the following speech rather than reading the tag aloud.[1], [2]
Speech generation in Timbre v3.0 supports nine Indian languages and Indian-accented English. Gnani lists Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Punjabi, Tamil and Telugu, with thirteen voices and thirteen audio tags across the ten languages. The company says a voice retains its character when switching languages. Text without tags also adapts delivery to meaning, tone and punctuation, leaving explicit cues for passages where the writer wants more control.[1]
Text normalisation processes amounts, dates, abbreviations and numbers before they are spoken. An account number read digit by digit calls for a different rendering from a currency amount spoken as a whole. Gnani identifies payment reminders, customer support, fraud alerts and appointment messages as business applications. It also names audiobooks, educational narration, game dialogue and video voiceovers among uses for the model’s delivery controls.[1]
