Uplift AI builds speech models for Pakistani languages so that voice agents, dubbing pipelines, and social-media content can speak and understand Urdu. The Orator text-to-speech family is the flagship; Urdu is the production language, with Sindhi and Balochi voices in the family and Punjabi and Saraiki announced as upcoming.
Text-to-speech is available as a REST endpoint (/v1/synthesis/text-to-speech) and as a WebSocket stream for conversational use, with about 300 ms latency. Voice profiles include an information/education voice, a broadcaster-style news voice, a storytelling voice, and a younger conversational voice; output formats include MP3 at 22.05 kHz. Speech-to-text currently covers Urdu only through two models, scribe and scribe-mini, with optional phone-commerce and farming domain settings; it transcribes segmented audio over HTTP rather than a continuous stream.
An AI calling preview places an outbound phone call and holds a conversation in Urdu; LiveKit voice-agent and WhatsApp bot guides are provided. A Pipecat STT service wraps the transcription endpoint with voice-activity-based segmentation.
Features
- Urdu text-to-speech: REST and WebSocket synthesis at roughly 300 ms latency
- Voice profiles: education, news broadcaster, storyteller, and contemporary conversational voices
- Regional languages: Sindhi and Balochi voices, with Punjabi and Saraiki in development
- Urdu speech-to-text: scribe and scribe-mini models with phone-commerce and farming domain options
- AI calling: an outbound Urdu voice-agent call demo
- Framework guides: LiveKit voice agents, WhatsApp bots, and a Pipecat STT service
- Client examples: cURL, JavaScript, and Python request samples