Smallest.ai Partners with Tenstorrent to Slash Voice AI Costs

Smallest.ai is now running its real-time text-to-speech model, Lightning V2, on Tenstorrent hardware. [2] The companies state that this text-to-speech system achieves production-quality audio under aggressive low-precision compute. [2] By utilizing this approach, Lightning V2 marks the first production-grade system of its kind to match the cost of text tokens without a degradation in audio quality. [2]

The partnership is designed to bring on-premises voice AI within reach for enterprises operating at any scale, achieving a four-fold reduction in costs through hardware acceleration. [1][2] According to the announcement, this move delivers 3.6 times lower infrastructure costs than leading GPU alternatives. [2] Furthermore, the hardware integration results in higher throughputs and faster response times. [2] The Lightning V2 system is available now, exclusively on Tenstorrent compute. [2]