SparkAudio/Spark-TTS

Features

• Autoregressive Transformers - Implements an autoregressive transformer architecture to generate high-fidelity speech by predicting audio tokens based on preceding context.
• Zero-Shot Voice Cloning - Enables zero-shot voice cloning to mimic target speaker identities using only short reference audio samples without requiring additional model training.
• Text-to-Speech - Converts written text into high-fidelity audio using advanced neural speech synthesis models.
• Cross-Lingual Speech Generators - Produces high-quality spoken output in multiple languages while maintaining consistent speaker identity.
• Generative Audio Engines - Provides a framework for synthesizing speech with adjustable parameters for pitch, speed, and vocal characteristics.
• Multilingual Speech Models - Generates natural-sounding audio across multiple languages while preserving unique speaker identity.
• Speaker Embeddings - Extracts acoustic features from short audio samples to condition synthesis models without additional training.
• Speech Processing - Efficient text-to-speech synthesis model.
• Speech Synthesis Models - Decodes compressed latent representations into high-fidelity audio waveforms using neural speech codec synthesis.
• Autoregressive Models - Generates coherent speech by predicting successive audio tokens based on preceding context.
• Prosody Control Tokens - Modulates speech pitch, speed, and prosody by manipulating latent vector dimensions during inference.
• Speech Synthesis Markup - Provides controls for adjusting pitch and speed to refine vocal output during synthesis.
• Speech Attribute Controls - Adjusts vocal characteristics like pitch and speaking rate through coarse or fine settings.

Star history