telli Duo
AI model · Multimodal model · Text → Text by telli
Full-duplex speech-to-speech AI for natural, human-like voice conversations.
telli Duo is a speech-to-speech AI model designed for natural, human-like phone conversations. It utilizes a full-duplex architecture that allows the agent to listen and speak simultaneously, handling interruptions and corrections in real time. The model is built for B2C businesses to automate high-volume outbound and inbound calls across sales, support, and operations. By combining a real-time speech model with a secondary reasoning model, it maintains conversational flow while executing complex tasks in connected business systems. It is delivered as part of the telli platform and supports ov
- Languages
- English, German, Spanish, French, Italian, Portuguese, Dutch
- Architecture
- Full-duplex speech-to-speech
- Latency
- Sub-second
- Supported Languages
- 40+
- Modality
- Audio to Audio
Overview
telli Duo is a speech-to-speech AI model designed for natural, human-like phone conversations. It utilizes a full-duplex architecture that allows the agent to listen and speak simultaneously, handling interruptions and corrections in real time. The model is built for B2C businesses to automate high-volume outbound and inbound calls across sales, support, and operations. By combining a real-time speech model with a secondary reasoning model, it maintains conversational flow while executing complex tasks in connected business systems. It is delivered as part of the telli platform and supports over 40 languages with sub-second latency.
Conversational Intelligence
- Full-duplex Interaction: The model listens and speaks at the same time, allowing for natural conversational rhythms including interruptions and corrections.
- Dual-model Architecture: A dual-model approach uses one model for real-time speech while delegating complex reasoning tasks to a secondary intelligent model.
- Native Interruption Handling: The model handles the flow of speech and interruptions directly without relying on traditional cascading voice activity detection.
- Task Delegation: It is prompted to delegate tasks that require deeper reasoning to another model to maintain conversation quality.
Voice Infrastructure
- Sub-second Latency: The model enables natural phone conversations with sub-second latency for fluid responsiveness.
- Multilingual Support: It supports over 40 languages, allowing businesses to engage global audiences with local accents and cultural context.
- Voicemail Detection: The model includes built-in voicemail detection to handle non-human answers appropriately.
- Warm Transfers: It supports warm transfers to human agents, including limited-hour transfer controls.
Specifications
- Architecture
- Full-duplex speech-to-speech
- Latency
- Sub-second
- Supported Languages
- 40+
- Modality
- Audio to Audio
Added by AntSpace from public sources (telli.com).