Gnani AI 发布 Timbre v3.0：AI 语音可低语、大笑与喊叫

Gnani AI has launched Timbre v3.0, a text-to-speech model that lets businesses and content creators control how AI-generated voices deliver speech, including pauses, whispers, laughter and changes in pace.

Timbre v3.0 also supports audio tags, which allow developers to specify how individual lines should be spoken. The model can generate its first audio in under 200 milliseconds, making it suitable for real-time voice conversations.

“Building speech AI for India means paying attention to how people actually speak and listen,” said Bharath Shankar, Chief Product and Engineering Officer at Gnani AI.

“A story needs pauses, changes in pace and expressions. With Timbre v3.0, we are bringing that clarity and control into one multilingual model, so businesses and creators can shape both what a voice says and how it says it,” he added.

Timbre v3.0 allows developers to insert audio tags into text to direct how a voice delivers specific lines. These include cues for laughing, giggling, sighing, whispering and shouting, as well as instructions to pause, slow down or speak with excitement.

For text without explicit audio tags, the model automatically adjusts its delivery based on the meaning, tone and punctuation of each sentence. The audio tags work across all supported languages.

The model is built to maintain a voice’s character across languages, allowing businesses to use a consistent voice for customer interactions across different regions.

It also includes text normalisation capabilities that convert written numbers, dates, currencies, symbols and abbreviations into spoken language. This is particularly relevant for banking, financial services and insurance applications, where voice agents need to clearly communicate payment amounts, account numbers, due dates and interest rates.

Developers can adjust speaking speeds to give listeners more time to understand important information or move faster through routine updates.

Gnani AI said Timbre currently powers more than one lakh hours of voice calls.

Timbre v3.0 supports streaming playback, allowing audio to begin before the model finishes generating the full response. It also supports long-form speech synthesis for extended narration and responses.

Enterprise use cases include payment reminders, collections, know-your-customer (KYC) processes, customer onboarding, insurance renewals, claims, appointment reminders and customer support across financial services, healthcare and telecom.

The model can also be used for audio articles, in-app guidance and accessibility tools that read digital content aloud in a user’s preferred language.