Amazon launches new advanced AI voice model Nova Sonic: What makes it special
Amazon has launched Nova Sonic, an advanced AI voice model, aims to outperform competitors in speed, accuracy, and conversational quality.
Amazon has introduced a new generative AI model, Nova Sonic, designed to process voice and produce speech that sounds more natural than previous digital assistants. The model, which Amazon claims rivals leading voice models from OpenAI and Google, focuses on speed, speech recognition, and conversational fluidity.
Nova Sonic aims to address the gap left by older AI voice models, such as Amazon Alexa and Apple Siri, which often sound stiff compared to newer technologies like ChatGPT's Voice Mode. These advancements have pushed earlier models into the background, highlighting the need for more adaptable and responsive AI interactions.
Also read: Oppo F29 Pro Review: Polished performer that delivers where it matters
Integration into Amazon’s Developer Platform
Available through Amazon's Bedrock platform, Nova Sonic offers a bi-directional streaming API to help developers integrate the voice model into enterprise AI applications. Amazon has highlighted the cost-efficiency of Nova Sonic, claiming it is about 80 percent more affordable than OpenAI's GPT-4o, a key competitor in the AI voice market. Components of the new model are already part of the upgraded Alexa+ voice assistant.
Key Strengths: Accuracy and Speed
In an interview with TechCrunch, Rohit Prasad, Amazon’s SVP and Head Scientist of AGI explained that Nova Sonic builds upon Amazon’s existing strengths in large orchestration systems, the underlying framework for Alexa. The model excels in directing user requests to the right APIs, enabling it to fetch real-time information or interact with external applications when necessary. Prasad emphasized that Nova Sonic can determine when to act and which tool to use based on the context of the conversation.
In its interactions, Nova Sonic listens for appropriate moments to speak, factoring in natural pauses and interruptions from the user. The model also generates a text transcript of the user’s speech, which developers can leverage in their own applications. According to Amazon, Nova Sonic is more adept than other models at understanding speech, even in noisy environments or when the user mumbles or misspeaks.
On the Multilingual LibriSpeech benchmark, which evaluates speech recognition across several languages, Nova Sonic achieved an impressive word error rate (WER) of 4.2 percent, meaning it was able to accurately transcribe speech in English, French, Italian, German, and Spanish 95.8 percent of the time. Additionally, on the Augmented Multi-Party Interaction benchmark, which measures accuracy during loud conversations with multiple participants, Nova Sonic outperformed OpenAI's GPT-4o model by 46.7 percent.
Also read: Infinix Note 50s 5G+ with ‘Energising Scent-Tech’ set to launch in India on April 18 - Details
Amazon’s Broader AGI Strategy
Amazon has also highlighted Nova Sonic’s speed, with an average latency of just 1.09 seconds. This places it ahead of GPT-4o’s Realtime API, which responds in 1.18 seconds, according to Artificial Analysis benchmarking.
Prasad noted that Nova Sonic is part of Amazon's broader goal to develop artificial general intelligence (AGI) - systems capable of performing tasks that humans can do on a computer. Future AI models from Amazon will expand beyond voice and text, incorporating sensory data like images and videos to create more dynamic interactions.
In line with this vision, Amazon recently previewed Nova Act, another AI model that integrates with Alexa+ and features like the Buy for Me service. Prasad shared that Amazon intends to offer more internal AI models to developers, encouraging innovation and the use of their technology in a wide range of applications.
