MCP Voice Engine

Deterministic, streaming-first prosody engine for expressive voice synthesis, pitch control, and real-time voice transformation.

Why this exists

Most voice DSP systems fail in two places: stability (warble, jitter, note flutter) and reproducibility (“it only happens sometimes”). MCP Voice Engine is built to be musical, causal, and deterministic—so it behaves like software, not folklore.

What you can build with it

• Real-time voice stylization for games and interactive apps (stable targets, expressive controls)
• Streaming voice pipelines (servers, bots, live processing)
• DAW / toolchain integration (deterministic pitch targets, consistent render behavior)
• Web Audio demos (AudioWorklet-ready architecture)

Quickstart

npm i
npm run build
npm test

Core capabilities

Deterministic output

Same input + config (and chunking policy) produces the same output, with regression protection via hash-based tests.

Streaming-first runtime

Stateful, causal processing designed for low latency. No retroactive edits. Snapshot/restore supported for persistence and resumability.

Expressive prosody controls

Event-driven accents and boundary tones let you shape cadence and intonation intentionally—without destabilizing pitch targets.

Meaning tests (semantic guardrails)

The test suite enforces communicative behavior, including:

• accent locality (no “smear”)
• question vs statement boundaries (rise vs fall)
• post-focus compression (focus has consequences)
• deterministic event ordering
• style monotonicity (expressive > neutral > flat without increasing instability)

Documentation

Primary docs live in packages/voice-engine-dsp/docs/.

Key documents

• Streaming Architecture
• Meaning Contract
• Debugging Guide
• Reference Handbook

Repository structure

packages/voice-engine-dsp/ — core DSP + streaming prosody engine, tests, and benchmarks

Running the test suites

npm test

Or run specific suites:

npm run test:meaning
npm run test:determinism
npm run bench:rtf
npm run smoke