AI Briefing
KO

MOSS-TTS-v1.5 Performance Improvements

·2026.05.27 00:32

Key point

MOSS-TTS-v1.5 is a TTS model with improved multilingual synthesis performance and voice cloning stability.

Details

MOSS-TTS-v1.5 inherits from the previous 1.0 version and provides zero-shot voice cloning, long-form speech generation, token-level duration control, multilingual synthesis, and code-switching capabilities.

Key improvements are as follows:

  • Enhanced multilingual synthesis: Performance is improved when language tags are specified, supporting a total of 31 languages.
  • Voice cloning stability: Increases speaker similarity and reduces cloning variability, ensuring consistency in generated results.
  • Improved long reference and short-form cloning: Works stably even when the reference audio is much longer than the target text.
  • Explicit pause control: Markers such as [pause 3.2s] allow precise pauses to be inserted at desired positions.
open-sourcemultilingualttsvoice-cloningmoss-tts
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

Google AI BlogGoogle Releases Gemini 3.8 Flash TTS: Voice Customization via Natural Language PromptsRedditConvai Releases Laya, a Non-Autoregressive Decision ModelRedditOpen-Weight CLM Released as Jev Alternative, Agent Decision-Making 4x FasterPyTorchKRTrillion Labs Releases GravityOCR, an OCR Model Applying Self-Speculative DecodingHacker NewsJevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs DetectionGeekNewsRootless Linux Sandbox 'Drop' Released for Isolating AI Agents and Third-Party Programs