Palabra AI launches fast real-time translation and TTS - Voice AI Space Barcelona

Background and Products

Palabra.ai is a research lab that specializes in real-time speech translation. Its product offerings include translation services for video calls, offline events, and video streams. While the company initially focused on speech-to-speech translation, it has developed in-house automated speech recognition (ASR) and text-to-speech (TTS) models optimized for real-time use cases.

New Streaming TTS Model

Palabra.ai is launching a streaming TTS model designed for low-latency applications. Key features of this model include:

• A time-to-first-audio latency of approximately 35 milliseconds.
• Support for 8 languages initially, expanding to 14 upon public release (including Spanish, French, Portuguese, Brazilian Portuguese, Chinese, and Hindi).
• Flat pricing with no concurrency or character limits.
• Cross-lingual voice cloning with "de-accenting," which allows a cloned voice to sound native in any supported target language.

Business Model and Use Cases

The company currently operates on a product-based business model, selling translation solutions for events, streams, and calls. It plans to offer API access directly through its platform and via infrastructure partners such as Agora and LiveKit. Notable adoption of the technology has occurred in the adult industry, where content creators use it to translate live interactions with viewers.

Technical and Operational Details

• Data Privacy: The company maintains a zero-data retention policy, meaning no user data or audio traces are stored on its servers.
• Development: Training the translation models took over a year and a half, leveraging a research team that transitioned from computer vision to voice AI.
• Model Selection: When evaluating TTS models, developers are advised to look at independent benchmarks (such as Coval) and consider how concurrency limits and pricing structures affect scalability.