Case Study: Building unique AI Voices for Voicebots with Talkie AI

Conversational AI is a rapidly growing field that is helping users use their voice to navigate the complexities of their everyday lives. One such leader in the conversational AI space is a proud customer of Resemble AI, Talkie.ai.

Talkie.ai is a flexible voicebot platform, which uses artificial intelligence and natural language processing to provide remote customer service 24/7. What makes them especially useful is the ability to create conversational experiences and virtual agents without writing a single line of code.

Through the course of their journey with us this past year, we asked them a few questions to best understand their use of our synthetic voice technology.

1. Why are synthetic voices important to you?

Talkie.ai helps our clients implement great conversational AI customer experiences. A big part of that customer experience lies in providing voicebots that sound natural and are easy for customers to interact with over the phone. The quality of the voice used by the voicebot makes an immediate impression with the caller. A synthetic voice allows our voicebots to handle dynamic components of a conversation seamlessly while enabling our clients to have a unique voice talent that represents their brand across all components of the conversation. It was especially great that Resemble AI supports the creation of broadcast quality audio (44 kHz).

2. How has Resemble AI been able to help you?

Resemble AI helped us create cloned synthetic voices for our voicebots using voice recordings. With Resemble AI, we could create a unique cloned voice for a specific customer that suited their brand voice and desired customer experience. 

We were also greatly helped by the Resemble AI customer support team. During our testing phase, we learned that a small delay at the start of calls was resulting in a higher than expected number of dropped calls. The Resemble AI team immediately helped us to solve this technical issue, reducing the delay, resulting in higher call conversion rates for our client. Resemble AI really went the extra mile to ensure we were able to deliver a high quality experience.

3. How does Talkie AI use synthetic voices in their product?

Talkie.ai uses synthetic voices in combination with static recordings to allow for a seamless voicebot experience. Typically, we use the synthetic voice to greet the customer by name using Resemble Fill. Rather than try to build an exhaustive library of name recordings, the synthetic voice enables us to handle this dynamic aspect of the call in a way that blends seamlessly with the static recordings by the same voice actor.

4. What impressed you the most about Resemble AI compared to other vendors you might have considered?

There were two areas that really stood out for us in working with Resemble AI. First, the quality of the synthetic voices was great. Solutions from other vendors either have a tendency to sound robotic or they lack the variety of voices to achieve a unique brand experience for clients. The second area that impressed us was the Resemble AI customer support. The Resemble AI customer support team were very helpful in resolving issues in a matter of hours, rather than days or weeks, as we might expect to experience with a larger vendor.

5. What do you think conversational AI will be like in 5 years?

This is a fast-developing area of technology and it is always difficult to pinpoint exactly what it will be like in the future. That said, we would expect that in five years time, we will be able to use synthetic voices to handle entire conversations, rather than depend on a combination of static recordings blended with dynamic voice synthesis. As the conversational AI improves, we would also expect voicebots to be able to understand even more customer intents with greater precision, respond to those intents more naturally and with a greater degree of variety. At the moment, the range of responses has to be programmed, resulting in some conversation loops that can become repetitive in outlier situations. Ultimately, we hope that we can create better voicebots faster without the same level of need for training data.