The xAI speech-to-speech technology, Grok Voice, is now accessible to developers via the fal.ai platform. This integration provides real-time audio-to-audio inference, replacing traditional multi-model pipelines with a single model architecture to reduce latency to approximately 0.70 seconds.
The API utilizes bidirectional WebSocket streaming for simultaneous audio input and output. It supports over 25 languages with native accent variations and offers 26 distinct voice options. Additionally, the system includes audio cloning capabilities for brand-specific voice profiles.
Developers can utilize tool-calling features, allowing agents to execute external functions like checking inventory or pulling account data during conversations. The technology is currently deployed within Starlink’s support and sales operations, processing more than 15,000 calls daily.
Pricing for the service on fal.ai is set at $0.00083 per second of audio, which equates to approximately $3 per hour of continuous interaction. This release is strictly developer-focused and does not include updates for the consumer Grok application.
