Skip to main content

Transcribers

Corrected 19 September 2026 against docs/user-guide/ (Part 3, section 3.3) and the phone’s Security & compliance screen: Deepgram is not offered, and there is no Engine tab.
Transcribers (ASR - Automatic Speech Recognition) are the “Ears” of your AI agent. They are responsible for turning human speech into text in real-time.

Key Performance Indicators

  1. WER (Word Error Rate): How accurately the engine recognizes words.
  2. Latency: How quickly it produces text from sound.
  3. Acoustic Robustness: How well it handles background noise, accents, and poor phone connections.

Supported Providers

ElevenLabs (the default)

Cloud speech recognition that detects the language by itself. The default for new agents.

Sarvam AI (Indian Regional Expert)

Optimized for the nuances of Indian accents and regional languages like Hindi, Odia, and Tamil. We recommend Sarvam for any deployment targeting the Indian market.

Whisper (hosted on Azure)

Also in the speech-recognition list.

Configuration in Movoice AI

Set the speech-recognition provider on the agent’s Settings tab on the web (Agents > Configure > Settings) or under Edit > Voice > Engines > Speech input on the phone. The phone’s Settings > Security & compliance > TRANSCRIPTION also has a switch that sends Indian-language audio, including Odia, to Sarvam. Some providers only handle some languages.