netease-youdao/EmotiVoice

Features

• Speech Synthesis - Synthesizes seamless spoken audio from a mix of Chinese and English text.
• Voice Cloning - Implements a framework for replicating specific speaker identities by training custom acoustic models on personal audio datasets.
• Joint Acoustic-Vocoder Training - Employs a jointly-trained acoustic-vocoder pipeline to produce high-fidelity audio with reduced artifacts.
• Expressive Synthesis - Uses style and emotional embeddings to control the timbre and expression of generated speech.
• Grapheme To Phoneme Conversion - Transforms Chinese text into phonetic representations via number normalization and pinyin conversion.
• Acoustic Model Trainers - Provides joint training for acoustic models and vocoders to ensure high-fidelity synthetic audio generation.
• Voice Model Trainers - Utilizes a deep learning architecture to align text with high-fidelity emotional expressions.
• Synthetic Speech Generation - Produces high-fidelity synthetic speech by replicating vocal characteristics based on specific speaker profiles.
• Voice Synthesizer Training - Processes custom audio datasets and transcriptions to train models on specific speaker characteristics.
• Phoneme-Based Pipelines - Implements a pipeline to transform raw bilingual text into phonetic representations for synthesis.
• Emotional Synthesis - Generates synthetic audio in English and Chinese with controllable emotional states like happiness or sadness.
• Bilingual Synthesizers - Acts as a bilingual synthesis engine processing mixed Chinese and English text into seamless audio.
• Emotional Modulation - Generates synthetic audio that conveys specific human emotions like happiness or sadness.
• Emotional TTS Engines - Generates synthetic audio in English and Chinese with controllable emotional states.
• Model Fine-Tuning - Supports improving emotional expression by adapting pre-trained synthetic voices using custom datasets and alignment.
• Speech Synthesis Services - Operates as a containerized web server exposing speech synthesis capabilities through an HTTP interface.
• Text-to-Speech - Provides a self-hosted web service via Docker for programmatic text-to-speech generation.
• Local Speech Synthesis - Allows users to generate synthetic speech locally via a desktop application without an internet connection.
• Training Dataset Preparation - Includes utilities to organize datasets and initialize model checkpoints specifically for voice model training.
• Synthetic Voice Design - Provides a library of diverse speaker identities and gender profiles to define the characteristics of generated speech.
• Command Line Interfaces - Provides a command line interface for generating synthetic audio from text.
• Speech API Hosting - Ships an HTTP interface to expose synthetic voice generation programmatically to external applications.
• Text-to-Speech Engines - Provides a system to convert written text into spoken audio via remote server requests.
• API Wrappers - Exposes the internal speech engine via a web server wrapper for remote programmatic use.
• Web Dashboards - Ships an interactive browser-based dashboard for performing text-to-speech synthesis without writing code.

Star history