VoiceMPP

Fully self-hosted Voice-as-a-Service via Tempo micropayments. Pay-per-request TTS and STT over the Machine Payments Protocol (MPP).

Zero external API dependencies. All inference runs locally using Piper TTS and whisper.cpp.

Live demo: https://voicempp.shelmail.xyz OpenAPI spec: https://voicempp.shelmail.xyz/openapi.json MPPscan: registered on mppscan.com

Endpoints

Method	Path	Price	Description
POST	/api/tts	$0.05/request	Text to WAV audio (paid via MPP)
POST	/api/stt	$0.01/request	Audio to text transcript (paid via MPP)
POST	/api/demo/tts	Free	TTS demo (500 char limit)
POST	/api/demo/stt	Free	STT demo (5MB limit)
GET	/api/voices	Free	List available TTS voices
GET	/api/health	Free	Service health check
GET	/openapi.json	Free	OpenAPI 3.1.0 discovery spec

Tech Stack

Component	Technology
TTS engine	Piper — ONNX neural TTS, CPU
STT engine	whisper.cpp — C++ Whisper, CPU
Payments	mppx SDK — Tempo USDC
Server	Node.js + TypeScript (raw HTTP, no framework)
Discovery	OpenAPI 3.1.0 + MPPscan compatible

Setup Guide

Prerequisites

• Node.js 20+
• ffmpeg (apt install ffmpeg)
• cmake + build-essential (for whisper.cpp)
• A Tempo wallet private key

1. Clone and install

git clone https://github.com/dieutx/voicempp.git
cd voicempp
npm install

2. Install Piper TTS

mkdir -p bin models
# Download Piper binary
wget -q https://github.com/rhasspy/piper/releases/download/2023.11.14-2/piper_linux_x86_64.tar.gz -O /tmp/piper.tar.gz
tar xzf /tmp/piper.tar.gz -C /tmp
cp /tmp/piper/piper bin/piper
cp /tmp/piper/lib* bin/ 2>/dev/null
cp -r /tmp/piper/espeak-ng-data bin/ 2>/dev/null

# Download voice models (add as many as you want from https://rhasspy.github.io/piper-samples/)
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/amy/medium/en_US-amy-medium.onnx -O models/en_US-amy-medium.onnx
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/amy/medium/en_US-amy-medium.onnx.json -O models/en_US-amy-medium.onnx.json
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/ryan/medium/en_US-ryan-medium.onnx -O models/en_US-ryan-medium.onnx
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/ryan/medium/en_US-ryan-medium.onnx.json -O models/en_US-ryan-medium.onnx.json

3. Build whisper.cpp

git clone --depth 1 https://github.com/ggerganov/whisper.cpp.git /tmp/whisper
cd /tmp/whisper && cmake -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j$(nproc)
cp build/bin/whisper-cli /path/to/voicempp/bin/whisper-cli
cd /path/to/voicempp

# Download Whisper model
wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin -O models/ggml-base.bin

4. Configure environment

cp .env.example .env

Edit .env:

PRIVATE_KEY=<your tempo wallet private key, hex>
MPP_SECRET_KEY=<any random hex string, e.g. openssl rand -hex 16>
PORT=3000

5. Run

# Development
LD_LIBRARY_PATH=bin npm run dev

# Production
npm run build
LD_LIBRARY_PATH=bin npm start

Docker (alternative)

cp .env.example .env
# Edit .env
docker compose up --build

Docker automatically downloads Piper, whisper.cpp, and all models.

Systemd service (production)

[Unit]
Description=VoiceMPP
After=network.target

[Service]
Type=simple
WorkingDirectory=/path/to/voicempp
Environment=LD_LIBRARY_PATH=/path/to/voicempp/bin
EnvironmentFile=/path/to/voicempp/.env
ExecStart=/usr/bin/node dist/server.js
Restart=on-failure

[Install]
WantedBy=multi-user.target

Usage

With mppx CLI

# TTS (paid)
mppx POST https://voicempp.shelmail.xyz/api/tts \
  -d '{"text":"Hello!","voice":"en_US-amy-medium","speed":1.0}'

# STT (paid)
mppx POST https://voicempp.shelmail.xyz/api/stt \
  -F file=@recording.wav

With agentcash

npx agentcash try https://voicempp.shelmail.xyz
npx agentcash fetch "https://voicempp.shelmail.xyz/api/tts" \
  -m POST -H "Content-Type: application/json" \
  -b '{"text":"Hello!","voice":"en_US-amy-medium"}' --max-amount 0.05

With mppx SDK (TypeScript)

import { Mppx, tempo } from "mppx/client";
import { privateKeyToAccount } from "viem/accounts";

const mppx = Mppx.create({
  methods: [tempo({ account: privateKeyToAccount("0x...") })],
});

// TTS
const ttsRes = await mppx.fetch("https://voicempp.shelmail.xyz/api/tts", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ text: "Hello!", voice: "en_US-amy-medium" }),
});
const audio = await ttsRes.arrayBuffer(); // WAV binary

// STT
const form = new FormData();
form.append("file", audioBlob, "recording.wav");
const sttRes = await mppx.fetch("https://voicempp.shelmail.xyz/api/stt", {
  method: "POST",
  body: form,
});
const { text, language, duration } = await sttRes.json();

TTS Parameters

Parameter	Type	Default	Description
text	string	required	Text to synthesize (max 5000 chars)
voice	string	en_US-amy-medium	Voice ID from /api/voices
speed	number	1.0	0.25 (slow) to 4.0 (fast)
noiseScale	number	0.667	Expressiveness (0=monotone, 1=expressive)
noiseW	number	0.8	Phoneme variation (higher=more natural)
sentenceSilence	number	0.2	Pause between sentences (seconds)
speaker	integer	0	Speaker ID for multi-speaker models

Adding Voices

Drop Piper ONNX models into models/ — auto-discovered on startup:

# Example: add German voice
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/de/de_DE/thorsten/medium/de_DE-thorsten-medium.onnx -O models/de_DE-thorsten-medium.onnx
wget -q https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/de/de_DE/thorsten/medium/de_DE-thorsten-medium.onnx.json -O models/de_DE-thorsten-medium.onnx.json
# Restart server to pick up new voice

Browse all voices: https://rhasspy.github.io/piper-samples/

Performance

Task	Latency	Notes
TTS (100 chars)	~250ms	15x faster than real-time
TTS (500 chars)	~500ms	Scales linearly
STT (5s audio)	~1.3s	whisper base model
STT (30s audio)	~5s	16 CPU cores

Payment Flow

Client                          VoiceMPP
  |                                |
  |--- POST /api/tts ------------->|
  |                                |
  |<-- 402 + WWW-Authenticate -----|  (payment challenge)
  |                                |
  |--- Pay USDC on Tempo --------->|  (on-chain transfer)
  |                                |
  |--- POST /api/tts + credential->|  (retry with proof)
  |                                |
  |<-- 200 + WAV audio ------------|  (done)

License

MIT