Deepdub launches Phantom Z 3.4 Conversational for improved multilingual text-to-speech

TEL AVIV, Israel, Deepdub, a leading voice AI company, has unveiled Phantom Z 3.4 Conversational, a new multilingual text-to-speech model designed to excel in real-world applications, not just demonstrations. This advanced model, with its high-fidelity 48 kHz audio and enhanced text normalization, aims to deliver seamless customer interactions.

In enterprise voice agent deployments, success hinges on four critical factors: the voice quality, response speed, timing of speech, and accuracy of spoken numbers, dates, and amounts. When one of these factors falters, the call often escalates to a human operator, affecting customer satisfaction and containment. Phantom Z 3.4 Conversational is crafted to address these challenges, ensuring that even the last few percent of accuracy are crucial.

The model's text normalization feature is particularly noteworthy. For instance, a date written as 2024-12-31 is pronounced as "December thirty-first, twenty twenty-four" instead of a string of numbers. Similarly, an invoice total of $1,240 is read as "one thousand two hundred forty dollars." These refinements are critical in ensuring that the spoken words align with the customer's expectations.

Phantom Z 3.4 Conversational excels in English, Hebrew, and other languages, thanks to Deepdub's proprietary speech models. The company's approach to building these models from scratch, rather than licensing them, allows for quick language integration. Deepdub currently supports over 50 locales and dialects, verified by local voice and language experts.

"Our deployment in real-time phone calls depends on the voice being both natural and accurate. 3.4 has brought us closer to a human-like interaction, significantly improving caller retention and engagement," said Adir Haziza, CTO of Voiceman. The company has seen a marked improvement in call duration and customer interaction.

Deepdub's latest model is particularly adept at handling Hebrew, a language where the absence of vowels can lead to ambiguity. Phantom Z 3.4 resolves this by determining the correct pronunciation based on the full sentence context, ensuring that every instance of a potentially ambiguous word is read correctly. In blind listening tests, Phantom Z 3.4 outperformed Deepdub's previous Hebrew model, with 71% of test cases showing a preference for the new model.

"We tested the model extensively to ensure it could handle the volume and variety of work we do. The standout feature was its accuracy and the naturalness of the Hebrew," said Dor Levy, Head of Jeen Talk at Jeen AI.

Phantom Z 3.4 Conversational is now available to all Deepdub clients, marking a significant advancement in the company's mission to provide cutting-edge voice solutions for global enterprises. By addressing the nuances of multilingual communication, Deepdub is poised to revolutionize the way businesses interact with their customers.