Sold by: Cartesia
Deployed on AWS
Cartesia Sonic delivers natural AI Voices in 44+ languages including accent localization and controls for emotional expressiveness, all at 2-4x lower latency than alternatives with industry leading reliability.
Overview
Cartesia is the only model where you don't have to pick between quality and speed. Our models are built on State Space Model (SSM) architecture, a fundamentally different approach to transformers, pioneered by our founding team at Stanford. For text-to-speech, this translates into three things that matter at production scale:
- Lowest latency in the market
- Lower cost and higher concurrency
- State-of-the-art naturalness
Highlights
- Sonic's support for 44+ language with accent localization and multilingual voices reaches customers around the world.
- Full control over emotional expressiveness, speed, volume and more, all at 2-4x lower latencies than alternatives.
- Achieve accurate pronunciation for complex phone numbers, addresses, and IDs every invocation.
Details
Sold by
Categories
Delivery method
Latest version
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on actual usage, with charges varying according to how much you consume. Subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost |
|---|---|---|
ml.m5.4xlarge Inference (Batch) Recommended | Model inference on the ml.m5.4xlarge instance type, batch mode | $0.001/host/hour |
inference.count.m.i.c Inference Pricing | inference.count.m.i.c Inference Pricing | $0.037/request |
Dimensions summary
You pay based on usage across two dimensions. The first bills the Sonic 3.6 model running in batch inference mode on the ml.m5.4xlarge instance type, charged per host hour. This covers the compute time the model runs while processing text-to-speech workloads. The second bills inference by request count, so you pay per inference call made. These dimensions work independently rather than as tiers. Host-hour billing scales with how long the instance runs, while request-based billing scales with call volume. You can use one or both depending on your deployment approach.
Top-of-mind questions for buyers
You pay for each hour the ml.m5.4xlarge instance runs the Sonic 3.6 model in batch mode. Batch mode processes text-to-speech workloads in bulk rather than per live request. The charge accrues while the instance is active. Powered-off instances stop the software host-hour meter, though underlying AWS resource fees may still apply.
The two dimensions bill independently and can appear on the same invoice. The host-hour charge scales with how long the ml.m5.4xlarge instance runs in batch mode. The per-request charge scales with the number of inference calls made. Batch processing favors host-hour billing; discrete real-time calls favor request-based billing.
Each inference call to the Sonic 3.6 model counts as one billable request. A request is a single text-to-speech generation sent to the model. Cost rises directly with call volume, so higher request counts increase this charge. This dimension bills per call rather than per hour of compute.
www.cartesia.ai
Helpful?
Vendor refund policy
For refunds, please contact support@cartesia.ai
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Amazon SageMaker model
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Version release notes
Initial Sonic 3.6 Launch; backwards compatible with Sonic 3.5
Additional details
Inputs
- Summary
The response streaming endpoint takes in a JSON object as the input that specifies the transcript, voice, language, and output format for the generation.
- Input MIME type
- application/json
{
"context_id": "0",
"transcript": "The detective burst through the door. 'We've got maybe five minutes before they realize we're here, so listen carefully and listen well: <speed ratio='1.5'/> the artifact is hidden beneath the old courthouse, exactly three feet below the cornerstone, and <volume ratio='0.5'/>whatever you do, DO NOT touch it with your bare hands!' She paused, catching her breath. 'Now... here's the important part... <speed ratio='0.6'/>you need to... very slowly... very carefully... wrap it in the copper wire first... then the silk cloth... then seal it in the lead box.' <volume ratio='2.0'/> Footsteps echoed in the hallway. 'GO GO GO! They're coming up the stairs RIGHT NOW!'",
"language": "en",
"output_format": {
"container": "raw",
"sample_rate": 44100,
"encoding": "pcm"
}
"voice_id": {
"mode": "id",
"id": "bf0a246a-8642-498a-9950-80c35e9276b5"
},
}
Do not use batch mode.
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name | Description | Constraints | Required |
|---|---|---|---|
context_id | A unique ID provided by the client to identify the request. It can be any string value and helps with tracking or debugging | - | Yes |
transcript | The text that will be converted into speech. You can include additional controls (e.g., emotion, speed, volume) as supported by Sonic 3 models: https://docs.cartesia.ai/build-with-cartesia/sonic-3/volume-speed-emotion | - | Yes |
language | The language code of the transcript text. Supported codes include: ar, bg, bn, cs, da, de, el, en, es, fi, fr, gu, he, hi, hr, hu, id, it, ja, ka, kn, ko, ml, mr, ms, nl, no, pa, pl, pt, ro, ru, sk, sv, ta, te, th, tl, tr, uk, vi, zh, en-US, de-DE, es-MX, fr-FR, hi-IN, it-IT, ja-JP, nl-NL, pt-BR | If not provided, the transcript language will be automatically detected | No |
output_format | Must match the raw option from the Cartesia TTS SSE API: https://docs.cartesia.ai/api-reference/tts/sse#body-output-format. Only raw is supported | - | Yes |
voice | Matches the voice field from the Cartesia TTS SSE API: https://docs.cartesia.ai/api-reference/tts/sse#body-voice. Only mode = id is supported. Example: { "mode": "id", "id": "voice_123" } | - | Yes |
generation_config | Optional configuration object matching the API schema: https://docs.cartesia.ai/api-reference/tts/sse#body-generation-config | - | No |
add_timestamps | Whether to include word-level timestamps in the output: https://docs.cartesia.ai/api-reference/tts/sse#body-add-timestamps | - | No |
add_phoneme_timestamps | Whether to include phoneme-level timestamps in the output: https://docs.cartesia.ai/api-reference/tts/sse#body-add-phoneme-timestamps | - | No |
use_normalized_timestamps | Whether timestamps should be normalized (0–1 range): https://docs.cartesia.ai/api-reference/tts/sse#body-use-normalized-timestamps | - | No |
pronunciation_map | Optional map of word-to-pronunciation pairs that customizes how specific words are spoken. Keys are the words as they appear in the transcript; values are the desired pronunciations. Use this to override default pronunciations for domain-specific terms, proper nouns, acronyms, or any word the model mispronounces: https://docs.cartesia.ai/build-with-cartesia/capability-guides/custom-pronunciations Example: {"Cartesia": "kar-TEE-zhuh", "API": "A-P-I"} | 16 KB max size | No |
Support
Vendor support
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Cartesia Sonic delivers natural AI Voices in 40+ languages including accent localization and controls for emotional expressiveness, all at 2-4x lower latency than alternatives with industry leading reliability.
Get started with state-of-the-art Sonic and Ink Voice AI models, plus usage of Line Voice Agents all under one simple subscription.
Purple AI Model Context Protocol (MCP) Server provides secure, seamless integration between the Singularity Platform and any AI framework or large language model. Acting as a universal translator and intelligence hub, it empowers developers and partners to build custom agentic AI experiences powered by the full context and analytics of the SentinelOne platform. The open-source Purple AI MCP Server is available today on GitHub and will also be deployable as an EKS and through Amazon Bedrock, using Agent Core. Users will need to have an active deployment of SentinelOne console and be able to obtain the SentinelOne Singularity Console token and url to be able to deploy and use the Purple AI MCP server.
P3 Secure AI provides secure and governed access to AI models, allowing users to experiment with, and adopt, AI models at the pace of mission. P3 Secure AI offers LLM flexibility and choice, avoiding LLM lock-in and provides as-code, AI service-specific guardrails to control access and manage costs.
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.