Deterministic · Streaming-first · Zero-copy PCM

Voice DSP that behaves like software, not folklore.

Streaming TTS inference backend with Kokoro/Piper synthesis, expressive prosody controls, and deterministic output. Powers mcp-voice-soundboard.

Quick start Read the Handbook

Install

npm i && npm run build

Test

npm test npm run test:meaning npm run test:determinism

Bench

npm run bench:rtf npm run smoke

Core Capabilities

Built for stability and reproducibility — the two places most voice DSP systems fail.

Deterministic Output

Same input + config + chunking policy produces identical output every time. Regression-protected via hash-based tests — not "it sounds about right".

Streaming-First Runtime

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

Expressive Prosody

Event-driven accents and boundary tones shape cadence and intonation intentionally. Meaning tests enforce accent locality, question vs statement boundaries, and post-focus compression.

Meaning Tests

The test suite enforces communicative behavior — not just "does it run".

Test

What it enforces

accent locality

Accents affect the target pitch region only — no smear into adjacent syllables

question vs statement

Rising boundary tone for questions, falling for statements — reliably distinct

post-focus compression

Focus events have consequences — compressed range after the focal word

deterministic event ordering

Same prosody event sequence always produces the same pitch curve

style monotonicity

expressive > neutral > flat — without increasing instability at higher expressivity

Packages

Monorepo — one package today, clean separation for future synthesis backends.

Package

Contents

packages/voice-engine-dsp

Core DSP + streaming prosody engine, meaning tests, determinism tests, RTF benchmarks

Quick Start

Build & test

git clone https://github.com/mcp-tool-shop-org/mcp-voice-engine
cd mcp-voice-engine
npm i
npm run build

# Full test suite
npm test

# Specific suites
npm run test:meaning       # communicative behavior
npm run test:determinism   # hash regression tests
npm run bench:rtf          # real-time factor benchmark
npm run smoke              # end-to-end smoke test

Key docs

packages/voice-engine-dsp/docs/
├── STREAMING_ARCHITECTURE.md  # causal processing model
├── MEANING_CONTRACT.md        # prosody behavior spec
└── DEBUGGING.md               # debugging guide

Reference_Handbook.md          # full API + concepts reference
PERF_CONTRACT.md               # RTF and latency guarantees