MeloTTS Deployment Guide

MeloTTS from MyShell is a compact multilingual TTS library with support for English, Spanish, French, Chinese, Japanese, and Korean. On our dedicated GPU hosting it is the fastest permissively-licensed option at reasonable quality.

Contents

• Install
• Minimal example
• Throughput
• Versus alternatives

Install

pip install melo-tts
python -m unidic download

Example

from melo.api import TTS

tts = TTS(language="EN", device="cuda")
speaker_ids = tts.hps.data.spk2id

text = "Hello from a GigaGPU dedicated server."
tts.tts_to_file(text, speaker_ids["EN-US"], "out.wav", speed=1.0)

Pre-shipped speakers: EN-US, EN-BR (British), EN-INDIA, EN-AU, EN-Default. Multilingual code supports ES, FR, ZH, JA, KR.

Throughput

On a 4060 8GB:

• 10-word sentence: ~0.3 seconds generation
• Real-time factor: ~0.05x (20x faster than playback)
• Throughput: hundreds of seconds of audio per minute

MeloTTS is notably faster than Parler-TTS or Bark because the model is smaller and the architecture is optimised for speed over voice variety.

Alternatives

For high-throughput voice needs (notifications, narration pipelines), MeloTTS wins on cost. For expressive voiceovers, Parler or Fish Speech sound better.

See Parler-TTS and Fish Speech.