What problem does it solve?
This Skill provides a complete, end-to-end guide for generating high-quality voice outputs, covering both singing (ACE-Step) and speech (Kokoro/Coqui TTS), along with practical routing and production-ready implementation patterns.
Core Features & Use Cases
- Singing with ACE-Step: Generate full songs with vocals and musical accompaniment, including lyric alignment and structure.
- Speech with TTS: Produce natural-sounding speech with Kokoro TTS or Coqui TTS for voice cloning and multilingual output.
- Routing & Production Patterns: Automatic model selection based on user intent (singing vs speech) and production-ready pipeline integration with metadata tagging.
Quick Start
Install the required Python packages, then run a quick example to generate a 60-second singing sample:
- Install dependencies:
pip install acestep kokoro soundfile numpy TTS
- Python snippet to generate a sample:
from acestep.pipeline_ace_step import ACEStepPipeline
pipeline = ACEStepPipeline(dtype='bfloat16', cpu_offload=True, overlapped_decode=True)
pipeline.load_checkpoint()
results = pipeline(
prompt='pop, upbeat, female vocal',
lyrics='[intro-short]\n\n[verse]\nYour topic here...\n\n[chorus]\nCatchy hook...\n\n[outro-short]',
audio_duration=60.0,
infer_step=27,
save_path='generated/song.wav'
)
The script will produce a WAV file and a corresponding JSON metadata file.
