StyleStream 2.0 cuts streaming voice conversion latency to 520 ms
Original title: StyleStream 2.0: Fast and Controllable Streaming Voice Style Conversion
A text-conditioned style embedding supports prompt-based voice design and instruction-based editing of one attribute.
StyleStream 2.0 is a streaming zero-shot voice style conversion system that replaces StyleStream 1.0's roughly 1-second latency and lack of a text interface. A few-step pixel MeanFlow model, fine-tuned with autoregressive feedback inspired by Self Forcing, reaches 520 ms end-to-end on a consumer GPU, about twice as fast, while keeping conversion quality. Target style is carried by a compact embedding and a unified text-conditioned flow-matching model for voice design and attribute editing.
- The paper reports the strongest target style fidelity on conversion, design, and editing.
- Intelligibility and preservation of non-edited attributes remain competitive.
- Mel-spectrogram context is removed so style is routed through the embedding.
- Latency is measured on a consumer GPU.
- Products
- StyleStream 2.0StyleStream 1.0
From the original article
1 min read at huggingface.co
Voice style conversion (VSC) aims to transform a source utterance to match the timbre, accent, and emotion of a target voice while preserving linguistic content. StyleStream 1.
By Yisi Liu, Nicholas Lee, Gopala K. Anumanchipalli · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.