Mobile

ConvoAI – AI Voice Call Assistant

ConvoAI is a voice assistant designed specifically for Android. It bridges the gap between traditional telephony and modern AI by using OpenAI's Whisper model to transcribe calls in real time, allowing users to interact through voice commands and intelligent summaries.

Project typeIndependent build
StatusCompleted project
FocusMobile

Technology

AndroidWhisperSpeech RecognitionKotlinOpenAI

System approach

  • Speech Recognition: OpenAI Whisper (on-device optimization).
  • Mobile Framework: Kotlin-based Android application with background services.
  • AI Processing: Integration with LLMs for intent classification and summarization.

Challenges

  • Managing audio stream latency for real-time transcription.
  • Balancing high-accuracy AI models with mobile battery life.
  • Ensuring privacy and security for sensitive call data.

Solutions

  • Implemented a chunked audio processing buffer to stream data to Whisper without waiting for silence.
  • Used quantized model versions to reduce the computational load on the mobile CPU.
  • Enabled end-to-end encryption for all processed audio fragments.

Key outcomes

01

Real-time speech-to-text

02

Low-latency intent detection

03

Battery-efficient processing