How miler777/omnivoice works
OmniVoice is a multilingual zero-shot text-to-speech system for speech generation, voice cloning, and voice design. The principal workflow accepts text and optional voice references or speaker instructions, prepares inputs, generates audio with the OmniVoice model, and returns speech. Separate branches cover model training and speech-quality evaluation. The graph includes documented capabilities whose implementation details were not sampled; dashed edges with null evidence mark relationships that cannot be verified from the excerpts.
Main entry points: omnivoice.py TTS model · builder.py Training builder · train.py Training CLI · wer WER evaluation
Components 24
Speech generation
- Python API
omnivoice/__init__.py - Web demo
omnivoice/cli/demo.py - CLI inference
omnivoice/cli/infer.py - Batch inference
omnivoice/cli/infer_batch.py - TTS model
omnivoice/models/omnivoice.py - FlashInfer model
omnivoice/models/omnivoice_flashinfer.py - Voice design
omnivoice/utils/voice_design.py
Model training
- Training CLI
omnivoice/cli/train.py - Training builder
omnivoice/training/builder.py - Training data
omnivoice/data/dataset.py - Batching
omnivoice/data/batching.py - Data collator
omnivoice/data/collator.py - Data processor
omnivoice/data/processor.py - Trainer
omnivoice/training/trainer.py - Training config
omnivoice/training/config.py
Speech evaluation
- WER evaluation
omnivoice/eval/wer - MOS evaluation
omnivoice/eval/mos/utmos.py - Speaker similarity
omnivoice/eval/speaker_similarity/sim.py
Shared processing
- Audio processing
omnivoice/utils/audio.py - Text processing
omnivoice/utils/text.py - Language mapping
omnivoice/utils/lang_map.py - Test-list reader
omnivoice/utils/data_utils.py - Evaluation audio
omnivoice/eval/utils.py
Other components
- User
Connections 32
- User → Python API: requests speech
README.md - Python API → TTS model: generates audio
README.md - User → Web demo: requests demo
README.md - Web demo → TTS model: generates with
omnivoice/cli/demo.py - User → CLI inference: requests inference
README.md - CLI inference → TTS model: runs inference
omnivoice/cli/infer.py - User → Batch inference: requests batch speech
README.md - Batch inference → TTS model: runs inference
omnivoice/cli/infer_batch.py - TTS model → Audio processing: processes audio with
omnivoice/models/omnivoice.py - TTS model → Text processing: processes text with
omnivoice/models/omnivoice.py - TTS model → Language mapping: uses language map
omnivoice/models/omnivoice.py - FlashInfer model → TTS model: extends model
omnivoice/models/omnivoice_flashinfer.py - TTS model → Voice design: supports voice design
omnivoice/models/omnivoice.py - User → Training CLI: starts training
- Training CLI → Training builder: builds training setup
omnivoice/cli/train.py - Training builder → Training config: uses configuration
omnivoice/training/builder.py - Training builder → TTS model: constructs model
omnivoice/training/builder.py - Training builder → Training data: builds data loaders
omnivoice/training/builder.py - Training builder → Batching: configures batching
omnivoice/training/builder.py - Training builder → Data collator: configures collator
omnivoice/training/builder.py - Training builder → Data processor: configures processor
omnivoice/training/builder.py - Training CLI → Trainer: starts training
omnivoice/cli/train.py - Trainer → TTS model: trains model
- User → WER evaluation: evaluates speech
- User → MOS evaluation: evaluates quality
- User → Speaker similarity: evaluates similarity
- WER evaluation → Test-list reader: reads test lists
omnivoice/eval/wer/fleurs.py - MOS evaluation → Test-list reader: reads test lists
omnivoice/eval/mos/utmos.py - Speaker similarity → Test-list reader: reads test lists
omnivoice/eval/speaker_similarity/sim.py - WER evaluation → Evaluation audio: loads audio with
omnivoice/eval/wer/minimax.py - MOS evaluation → Evaluation audio: loads audio with
omnivoice/eval/mos/utmos.py - Speaker similarity → Evaluation audio: loads audio with
omnivoice/eval/speaker_similarity/sim.py