Modelpublic

ResontraAI/voxedge-model-index

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes

Model Card

VoxEdge model index

This is the canonical public package index for VoxEdge, an offering from Resontra AI. VoxEdge evaluates and packages speech and agent models for bounded, low-latency execution on consumer CPUs and GPUs.

The machine-readable candidate inventory is voxedge-candidates.json. A row is not a performance or availability claim: only an entry with a 40-character Hub revision has a published immutable package. Planned and blocked entries expose the roadmap without publishing unreviewed weights.

Current publication set

Candidate	Task	State	Intended artifact repository
Qwen3 0.6B INT4 ORT	Local agent LLM	Published at `b5fd7962f198470f34e68bf30c6474886a6891cc`	`ResontraAI/voxedge-qwen3-06b-int4-ort`
Pocket TTS English INT8 ONNX	Compact streaming TTS	Pinned; benchmark work pending	`ResontraAI/voxedge-pocket-tts-english-int8-ort`
VoXtream2	CUDA streaming TTS comparison	CUDA benchmark pending	`ResontraAI/voxedge-voxtream2-cuda`
StreamVoiceAnon INT8 ORT	Voice conversion/anonymization	Local graphs verified; release gates open	`ResontraAI/voxedge-streamvoiceanon-int8-ort`
RVC v2	Voice conversion	Authorized checkpoint and components missing	`ResontraAI/voxedge-rvc-v2-ort`
Beatrice v2	Voice conversion	Provider runtime/terms pending	`ResontraAI/voxedge-beatrice-v2-runtime`

Safety and provenance boundary

Public VoxEdge repositories never contain prompt recordings, target identities, speaker embeddings, FAISS speaker indexes, consent records, runtime caches, generated review audio, secrets, or checkpoints without explicit redistribution rights. Speaker-specific packages are private or gated and separately deletable.

Benchmark claims remain tied to their measured hardware and immutable artifact hashes. Cloud results do not substitute for physical consumer-device evidence.

Source, schemas, and reproducibility tooling live in the VoxEdgeLab repository.