AdaSpeech Inference Guide

This guide explains how to properly run inference using the AdaSpeech model.

Installation

1. Clone the repository:

   git clone git@github.com:Cocii/AdaSpeech.git
   cd AdaSpeech

2. Install the required Python packages:

   pip install -r requirements.txt

Model and Vocoder Checkpoints

Before running inference, you need:

1. AdaSpeech model checkpoint
2. Vocoder checkpoint (default BigVGAN)
3. Configuration files:
   • preprocess.yaml
   • model.yaml
   • train.yaml

Running Inference

Basic Command Structure

CUDA_VISIBLE_DEVICES=0 python inference.py \
--language_id <LANG_ID> \
--speaker_id <SPEAKER_ID> \
--reference_audio <REF_AUDIO_PATH> \
--text "$(cat test.txt)" \
-p <PREPROCESS_CONFIG> \
-m <MODEL_CONFIG> \
-t <TRAIN_CONFIG> \
--restore_step <CHECKPOINT_STEP> \
--vocoder_checkpoint <VOCODER_PATH> \
--vocoder_config <VOCODER_CONFIG>

Key Parameters

• language_id: Language identifier (0 for English, 1 for Chinese)
• speaker_id: Target speaker identifier
• reference_audio: Path to reference audio file for speaker embedding
• text: Input text for synthesis (can be read from file)
• restore_step: Checkpoint step number to load
• vocoder_checkpoint: Path to vocoder model weights
• vocoder_config: Path to vocoder configuration file