rioharper/VocalForge

Combines Whisper transcription, PyAnnote speaker diarization, and CTC segmentation to automatically process raw audio into aligned speech datasets with minimal manual curation. The toolkit handles speaker isolation, voice activity detection, noise filtering, and text normalization across multiple audio sources, then exports in LJSpeech format. Includes VCAuditor, a browser-based verification interface for reviewing waveforms, correcting alignments, and filtering low-confidence segments before final dataset export.

130 stars. No commits in the last 6 months.