Researchers have developed ProPS, a novel framework for synthesizing speaker embeddings conditioned on natural language prompts. This system converts textual descriptions of speaker profiles into sentence embeddings, which then guide a mixture density network to predict Gaussian mixture models in the x-vector space. ProPS has demonstrated its ability to generate distributions of speaker embeddings that accurately reflect requested attributes such as age, gender, accent, and prosody, making it valuable for controllable speech generation systems like Text-To-Speech and Voice Conversion. AI
IMPACT Enables more controllable and nuanced voice generation for applications like TTS and voice conversion.
RANK_REASON The cluster contains a research paper describing a new framework for synthesizing speaker embeddings.
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gaussian mixture model
- Gotit.pub
- Hugging Face
- Prompted Profile Synthesis
- ScienceCast
- speech synthesis
- Voice conversion training method and server and computer readable storage medium
- X-Vectors: New Quantitative Biomarkers for Early Parkinson's Disease Detection From Speech
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
