Meet Nova Sonic, Amazon’s new AI voice model

AI companies have been working on voice models for a while now, but it seems things really ramped up after OpenAI unveiled ChatGPT Voice Mode.

Now, Amazon has just introduced its new “foundation” AI voice model called Nova Sonic. And it really makes Alexa sound like she’s living way in the past.

According to Amazon, Nova Sonic “unifies speech understanding and speech generation into a single model, to enable more human-like voice conversations in AI applications.” With the samples provided, it certainly does seem more human-like than the company’s previous iterations of AI voice models. 

For example, there are proper pauses, tone, and inflections on words depending on where they are and what they mean in a sentence. Amazon provided some samples you can listen to here and here.

Again, “more human-like” is the key description here. There are still plenty of signs that it’s an AI voice, but it also does sound like a big step over previous AI voice assistants like Alexa.

Amazon says that it achieved this by combining multiple models that would traditionally be used, like speech recognition, large language models, and text-to-speech, into one single unified model. According to Amazon, it not only understands the nuances in speech to produce it, but it also understands it when a human inputs their own speech with these nuances as well.

According to TechCrunch, Nova Sonic is already powering Amazon’s next-generation AI voice assistant, Alexa+.

Based on recent developments, it does seem like the big AI companies are currently focusing on voice models. So, prepare for competition in that space to heat up. Amazon is already pointing to claims that Nova Sonic is roughly 80 percent cheaper than OpenAI’s GPT-4o model and promoting it as “the most cost-efficient.”

Nova Sonic is currently available to developers through Amazon’s enterprise AI developer platform, Bedrock.