Scaled Cognition launches Voice API for accurate enterprise text-to-speech
The new model improves pronunciation of complex data like addresses and codes while offering lower costs than competitors through API or self-hosted options.
Key points
- Scaled Cognition introduced Voice, a text-to-speech model targeting enterprise customer service accuracy and cost reduction.
- The product supports over 25 languages, voice cloning, and integration with existing contact center infrastructure.
- Pricing starts at $9 per million characters via API, with self-hosted options costing significantly less at high volumes.
Enterprise customer service vendor Scaled Cognition Inc. launched a new text-to-speech model called Voice, designed to deliver precise pronunciation of account numbers, addresses, and other transactional details. According to SiliconANGLE, the company states the model achieves higher accuracy at a lower cost compared to competing services.
The Voice product is available through an application programming interface or for self-hosted deployment. It supports more than 25 languages and includes voice cloning capabilities that create digital replicas of specific voices. The system integrates with existing artificial intelligence agents, contact center platforms, and older telephone systems.
Context and Background
The launch addresses a gap between natural-sounding speech synthesis and the reliability required for customer service transactions. SiliconANGLE reported that modern generative models often struggle with alphanumeric confirmation codes, dollar amounts, and street addresses. These difficulties are particularly pronounced in languages other than English, according to the outlet.
Akshay Kannan, vice president of product management at Scaled Cognition, noted in e-mailed comments that competitors often stumble on these critical details. He stated that Voice enables enterprises to obtain natural-sounding speech and accurate readings without paying premium prices. The product aims to bridge the divide between audio quality and functional reliability.
Who is Affected
Scaled Cognition offers Voice at $9 per million characters through its hosted API. For self-hosted deployments, the software cost drops to $1 per million characters, excluding infrastructure expenses. Kannan explained that one graphics processing unit can handle approximately 32 simultaneous conversations using the model.
At scale, self-hosting becomes more economical. A redundant setup using two GPUs on standard cloud hardware costs roughly $6,000 a month and supports about 500 million characters. Including infrastructure, Kannan estimated costs at about $13 per million characters at that volume, dropping below $7 at 1.5 billion characters. He compared this to $35 to $100 per million charged by unnamed competitors.
What should you do?
Organizations evaluating text-to-speech solutions should assess their current volume against the self-hosting threshold. Teams with existing infrastructure may find self-hosting more cost-effective, while those without may prefer the hosted API. Security professionals should review the integration points for voice cloning features to ensure proper access controls and data handling protocols are in place.
What to do and how to stay safe: Scaled Cognition
- Review current text-to-speech usage volumes to determine if self-hosting offers a cost advantage over hosted APIs.
- Verify that voice cloning integrations comply with internal data privacy and consent policies.
- Monitor the performance of the model with alphanumeric codes and addresses in supported languages.
- Ensure legacy telephone systems are compatible with the new API before full deployment.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How much does Scaled Cognition's Voice model cost?
The hosted API costs $9 per million characters, while self-hosting software costs $1 per million characters, excluding infrastructure.
Does Voice support languages other than English?
Yes, the model supports more than 25 languages, according to SiliconANGLE.
Can Voice integrate with older telephone systems?
Yes, the product can work with existing AI agents, contact center platforms, and older telephone systems.
