Camb-ai/MARS5-TTS

Health v2 · maintenance only

Adoption not part of the score

2817 stars · 246 forks observed · 2026-08-28

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

MARS5 is an open-source English text-to-speech model from CAMB.AI that uses a two-stage AR-NAR pipeline to generate expressive speech with strong prosody. It supports voice cloning from a short (2-12 second) reference audio clip, with optional 'deep clone' mode using a reference transcript for improved quality.

Use cases

• clone a voice from a short audio sample
• generate expressive speech with natural prosody
• synthesize sports commentary or anime-style speech
• control pauses and emphasis via punctuation and capitalization
• run text-to-speech locally in Python or Colab

When to choose

• you need high-quality English TTS with expressive prosody
• you want open-source voice cloning from a few seconds of reference audio
• you want fine-grained prosody control through text formatting

When to avoid

• you need multilingual speech synthesis beyond English
• you need a production-grade hosted API with SLAs
• you cannot run GPU inference or accept AGPL-3.0 licensing

Facets

library · maturity active

tts speech-recognition machine-learning deep-learning speech-processing machine-learning python cross-platform voice-cloning text-to-speech prosody speech-synthesis ar-nar-pipeline english-only audio gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Camb-ai/MARS5-TTS")