Amazon Nova Sonic
Amazon · Voice & Language
Amazon Nova Sonic is a speech-to-speech model on Amazon Bedrock for voice AI that hears and answers in one pass, so you build a real-time voice assistant with low latency and live turn-taking.
About Amazon Nova Sonic
What Is Amazon Nova Sonic
Amazon Nova Sonic is Amazon's speech-to-speech foundation model, available through Amazon Bedrock. Instead of chaining a separate speech recognizer, language model, and text-to-speech engine, it handles the whole loop in one model and streams audio in both directions. That single-model design is what keeps the delay short enough for a live conversation.
The model targets developers who want a conversational AI agent that sounds human. It adapts to the rhythm of the person speaking, picks up on pauses, and lets the user cut in while it's still talking. You reach it through a bidirectional streaming API, so your app sends microphone audio and receives spoken replies over the same open connection.
The catch is the audience. Nova Sonic isn't a consumer app you open in a browser. It's infrastructure, sold through AWS, and it expects you to write code, manage an AWS account, and pay per token. It runs in supported AWS regions, so check availability for your area before you build.
Getting Started
- Sign in to an AWS account with access to Amazon Bedrock, then request model access for Nova Sonic in the region you plan to use.
- Install the AWS SDK for your language and pull one of the Python, Java, or NodeJS samples from the official Nova Sonic GitHub repo.
- Open a bidirectional stream, send 16 kHz mono audio from the microphone, and read the model's audio and text events as they arrive.
- Wire the output to your speaker or telephony stack, then add tool calling if the agent needs to fetch data or trigger actions.
Product Information
A quick look at Amazon Nova Sonic's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- Developers building voice agents
- Contact center teams
- Product teams testing voice features
Tasks
- Real-time voice conversations
- Speech translation and transcription
- Tool-driven voice workflows
Scenarios
- Customer service hotlines
- Hands-free field or in-car use
- Multilingual voice apps
Key features
Speech-to-Speech in a Single Model
Nova Sonic listens and speaks in one pass. You skip the usual three-part stack of recognition, reasoning, and synthesis, and the audio never leaves the model between steps. The result is a shorter delay and fewer moving parts to maintain.
Low-Latency Streaming
The bidirectional streaming API keeps a live channel open in both directions. Audio flows in and out continuously, so an answer starts before the user finishes a thought. That's the difference between a conversation and a walkie-talkie.
Natural Turn-Taking and Barge-In
The model reads pauses and adapts to how someone speaks. When a user cuts in mid-sentence, it stops and listens instead of talking over them. This back-and-forth is what makes a voice agent feel less robotic.
Tool Use and Knowledge Grounding
The agent can call external tools and pull from a knowledge base during a conversation. A spoken question can trigger a lookup, a booking, or a data fetch, and the answer comes back as speech. You can also log chat history to keep context across a session.
Built-In Speech Transcription
Nova Sonic returns text alongside audio, which covers transcription and captioning without another model. That saves money and keeps a single bill for the voice loop.
Integration with Voice Platforms
Nova Sonic works with LiveKit and Pipecat, plus Twilio and Vonage for telephony. That range means you can add voice to an app or a phone line without building your own streaming infrastructure.
Pros and cons
Pros
- One model covers recognition, reasoning, and speech output, so there's less glue code to write and maintain.
- Low latency and barge-in support make conversations feel natural rather than turn-based.
- Tool calling and knowledge base grounding let the agent take real actions mid-call.
- Support for English, Spanish, French, Italian, and German widens the markets you can serve.
- LiveKit, Pipecat, Twilio, and Vonage integrations shorten setup for voice and telephony apps.
Cons
- There's no free plan, and cost is measured per token, so a busy voice app adds up fast.
- It's a developer-only product sold through AWS, which shuts out anyone without coding skills or an account.
- Availability is limited to select AWS regions, so you may need to route traffic to a supported one.
- You handle the audio pipeline, credentials, and scaling yourself, which is real work beyond the model.
Frequently asked questions
It powers real-time voice agents that listen and speak in one pass. Typical uses include customer service lines, hands-free assistants, and speech translation.
Related content
Explore related tools, skills, and articles for Amazon Nova Sonic.
Amazon Nova Sonic Alternatives
Foley
Hanamori Labs, LLC · Voice & Language · Productivity · AI AssistantFoley is a voice to-do list that turns a spoken sentence into a filed task, complete with project, date and time, then helps you plan the day around what is already there.
HireKey
HireKey · Voice & Language · Productivity · LeaningHireKey is an AI job search copilot built by a former tech recruiter. It checks role fit, tailors resumes, runs interview practice and tracks every application.
Fluence
Yay Software (Yannick Huchard) · Writing · Voice & Language · ProductivityFluence is an Android app that reads PDFs, EPUBs, Word files and web articles aloud, follows each sentence as it speaks, and answers questions about the text on the phone itself.