Gnani AI adds tone and pace controls to its multilingual speech model Timbre v3.0

Gnani AI introduced Timbre v3.0, enabling businesses to control AI voice parameters like pace and expression. The model supports text-to-speech across nine Indian languages and Indian-accented English. Features include audio tags for nuances and text normalisation for clear communication of information. This speech generation technology is applicable in various sectors such as finance, healthcare, and telecommunications.

Bengaluru-based Gnani AI on October 14 launched Timbre v3.0, a speech-generation model that lets businesses and content creators control how an AI voice delivers a line, including its pace, pauses and expression, the company said.

Built in-house, the model converts text into speech across nine Indian languages and Indian-accented English. Gnani is targeting uses ranging from payment reminders and customer support calls to audiobooks and video voiceovers.

"Building speech AI for India means paying attention to how people actually speak and listen," Bharath Shankar, chief product and engineering officer at Gnani AI, said.

"Timbre v3.0 brings clear pronunciation, voices that retain their character across languages and control over expression at the moments that matter. Today over 1 lakh hours of voice calls are powered by Timbre", Shankar said.

The new version of the model introduces audio tags or short instructions in the text. Users can direct a voice to whisper, laugh or sigh, add a pause or change its speaking speed, the company said.

Gnani said these controls work across all supported languages. When users do not add instructions, the model adjusts its delivery based on the meaning, tone and punctuation of the text, it said.

The supported languages are Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Punjabi, Tamil and Telugu, in addition to Indian-accented English. Gnani said the model offers multiple male and female voices and retains a voice's characteristics when switching between languages.

For business applications, the company is emphasising the model's ability to turn written numbers, dates, currencies and abbreviations into spoken language. This process, known as text normalisation, helps distinguish between an amount that needs to be read as a sum and account digits that need to be spoken individually.

"A payment reminder needs to read an amount clearly. An insurance message needs to give the listener time to absorb a date. A learning app needs a voice that can hold attention through an explanation," Shankar said.

Gnani AI said Timbre v3.0 produces its first audio in under 200 milliseconds and starts playback before it finishes generating the entire output. It also supports longer narration, enabling use across phone systems, applications and digital content, the company said.

The company listed enterprise uses including customer support, identity verification, onboarding, insurance renewals and appointment reminders across financial services, healthcare and telecom.

Beyond business calls, the company is targeting audio articles, learning applications and tools that read content aloud in a listener's preferred language.