Skip to content

Bedrock Nova Sonic

Amazon Nova Sonic brings expressive, multilingual voice conversations to your agents. Users can speak naturally, interrupt a reply, and ask the agent to take action through tools.

Follow the quickstart to build a voice agent that listens, speaks, and calls tools.

Installation

Section titled “Installation”

Install the Nova Sonic extra:

Terminal window
pip install "strands-agents[bidi]"

For local audio dependencies and device setup, see Audio I/O.

Credentials

Section titled “Credentials”

Configure AWS credentials with permission to invoke Nova Sonic in a supported region. BedrockNovaSonicModel uses Boto3’s credential chain, including environment variables, named profiles, and IAM roles. See Setting up AWS credentials for shared setup instructions.

To use a named profile, pass a Boto3 session with its region:

import boto3


from strands.bidi.models import BedrockNovaSonicModel


model = BedrockNovaSonicModel(
    model_id="amazon.nova-2-5-sonic",
    boto_session=boto3.Session(profile_name="voice-agent", region_name="us-east-1"),
)

Without a custom session, you can pass region directly. If omitted, the model uses Boto3’s configured region, falling back to us-east-1.

Configuration

Section titled “Configuration”

Create a model with the voice and audio settings for your application, then pass it to BidiAgent. This example selects the tiffany voice and 24 kHz output audio:

from strands.bidi.agent import BidiAgent
from strands.bidi.models import BedrockNovaSonicModel


model = BedrockNovaSonicModel(
    model_id="amazon.nova-2-5-sonic",
    region="us-east-1",
    voice="tiffany",
    audio={"output": {"sample_rate": 24000}},
)
agent = BidiAgent(model=model)

Choose a voice from Nova Sonic’s voices and languages. Audio uses mono PCM, with 16 kHz input and output by default. Set either stream’s sample rate through audio.

See the BedrockNovaSonicModel API reference for all constructor options.

Session settings

Section titled “Session settings”

Pass Nova Sonic’s session settings through params to tune inference and turn detection. For example, LOW endpointing sensitivity allows longer pauses before the model decides the user has finished speaking:

from strands.bidi.models import BedrockNovaSonicModel


model = BedrockNovaSonicModel(
    model_id="amazon.nova-2-5-sonic",
    params={"turnDetectionConfiguration": {"endpointingSensitivity": "LOW"}},
)

Use the provider’s field names inside params, as shown above.

Connection restarts

Section titled “Connection restarts”

Nova Sonic limits each connection to eight minutes. By default, BidiAgent schedules a restart after seven minutes and replays conversation history into the new connection. See Connection restarts to customize the timing and track the transition.

On startup and restart, the adapter replays text from the conversation history, including completed transcripts. It truncates individual messages to 50 KiB and retains up to 200 KiB of recent history. These limits apply to the history sent when opening a connection.