Building a Tamil Voice Companion: Stack Choice Between Sarvam and Google, Long Memory, and Scaling
intrepidkarthi · reddit · 2026-08-20
The author is building a Tamil voice companion for long 5-10 minute conversations. Google Chirp3 HD offers better naturalness than Sarvam bulbul but lacks pitch adjustment and custom pronunciation. A cascade architecture (STT-LLM-TTS) is required over speech-to-speech to implement safety gates and memory.
Key questions discussed:
- Stack Selection: Seeking Tamil STT/TTS options (ElevenLabs and Deepgram are excluded due to lack of support or poor WER).
- Memory: Using structured extraction to SQLite (instead of RAG) to keep the prompt cache warm; asking for experiences with Graphiti, Zep, or Mem0 for non-English agents.
- Scaling: Comparing self-hosted LiveKit Agents vs. Pipecat. Current end-to-end latency is 2s, with plans to move to streaming WebSockets.
Related event: Devs Debate Tech Stack for Tamil Voice Companion App(2 posts)→
More from coding & agent
- Hermes Bot Mode test: research→implementation→verification handoff, 10/10 tests pass — Teknium · 2026-08-20
- Discussing Native Windows AI Coding Tools and Open Model Support — JadedSession · 2026-08-20
- Karpathy Principles for Claude Code: CLAUDE.md to Reduce Assumptions and Bloat — tom_doerr · 2026-08-20
- Useful Pattern: Vision OCR for Cells, Deterministic Logic for Structure — andrejusb · 2026-08-20
- Using Devin Daily for Two Months: Resetting Expectations on Productivity — Scobleizer · 2026-08-20
- shadcn echoes aidenybai: mainlining 9+ AI coding tools including Codex and Claude Code — shadcn · 2026-08-20