Freya

Freya develops speech models for text-to-speech, speech recognition and voice interaction, and provides voice agents that handle inbound and outbound phone calls. The platform addresses the need for conversational voice systems that can retrieve data from a company's own systems during a call and deliver results back afterward. Its English voice model, Freya TTS Adam V1, is presented alongside a blind listening benchmark in which listeners choose which of two clips sounds more human.

Core capabilities include a workflow builder for designing call flows without code, a simulation suite for testing agents against synthetic scenarios, and a dashboard for monitoring calls, recordings and outcomes. Integration happens through four connection points: a SIP phone line, a REST API for text-to-speech and speech-to-speech endpoints, mid-conversation HTTP requests to your own services, and a post-call webhook. Live transcription is available over WebSocket with Bearer key authentication. Deployment is offered on-prem, including air-gapped networks, or on cloud infrastructure in Türkiye, with role-based access, audit trails and data residency controls.

Freya also releases open research models, including FreyaTTS-small for Turkish speech synthesis, distributed under Apache 2.0 with code, model weights and an evaluation dataset. Pricing is determined per engagement based on call volume, deployment mode, agents and flows, models in use, integration work and forward-deployed engineering, with no published price list. The playground offers a free daily character allowance for generating speech without sign-in, with API keys available after signing in.