CSTR-Edinburgh/merlin

Health v2 · maintenance only

Adoption not part of the score

1321 stars · 431 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Merlin is a toolkit from the University of Edinburgh's CSTR for building deep neural network models for statistical parametric speech synthesis (text-to-speech) and voice conversion. It is written in Python (Theano-based, with optional TensorFlow/Keras support) and ships Kaldi-style recipes, requiring an external text-processing front-end and vocoder.

Use cases

• build a neural network text-to-speech voice
• train a statistical parametric speech synthesis model
• create a custom voice from my own recordings
• do research on deep learning for speech synthesis
• convert one voice to another with voice conversion
• run a demo TTS voice like slt_arctic

When to choose

• you need a proven, research-grade DNN-based TTS training toolkit with recipes
• you want an Apache-2.0 licensed system for academic or commercial speech synthesis research
• you are following tutorials or coursework on neural TTS voice building

When to avoid

• you need a modern, actively maintained framework - development has stalled and it relies on the legacy Theano library
• you want a plug-and-play TTS API rather than training your own models
• you need Windows support - it targets UNIX-like systems

Facets

library · maturity maintenance

machine-learning deep-learning speech-recognition tts audio-processing speech-processing machine-learning deep-learning python cross-platform text-to-speech speech-synthesis voice-conversion theano statistical-parametric-synthesis vocoder kaldi-style-recipes audio linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="CSTR-Edinburgh/merlin")