Ivan

Model · Audio classification

WeSpeaker-ResNet34-LM-MLX

Ivan

MLX-compatible weights for WeSpeaker ResNet34-LM, converted from the pyannote speaker embedding model with BatchNorm fused into Conv2d. WeSpeaker ResNet34-LM is a speaker embedding model (~6.6M params) that produces 256-dimensional L2-normalized speaker embeddings from audio. Trained on VoxCeleb for speaker verification and diarization. BatchNorm is fused into Conv2d at conversion time — no BN layers in the MLX model. Part of speech-swift. Converts the original pyannote/wespeaker-voxceleb-resnet34-LM checkpoint using a custom unpickler (no pyannote.audio dependency required). Key transformations: - Fuse BatchNorm into Conv2d: wfused = w × γ/√(σ²+ε), bfused = β − μ×γ/√(σ²+ε) - Transpose…

Model · Audio classification

Silero-VAD-v6.2.1-MLX

Ivan

MLX port of snakers4/silero-vad tag v6.2.1 for voice activity detection on Apple Silicon. Measured with speech-swift release tests on Apple Silicon using a 20 s 16 kHz speech fixture, 625 streaming chunks. Parity against the matching CoreML v6.2.1 export: The exported safetensors were also checked tensor-by-tensor against the upstream v6.2.1 JIT state dict after conversion; the maximum absolute tensor difference was 0. Converted from snakers4/silero-vad tag v6.2.1. The upstream project is MIT licensed. - speech-swift - Apple SDK - Docs - install and CLI docs - soniqo.audio - website - blog - blog

Model · Audio classification

Qwen3-ForcedAligner-0.6B-4bit

Ivan

4-bit quantized version of Qwen/Qwen3-ForcedAligner-0.6B for Apple Silicon inference via MLX. Predicts word-level timestamps for audio+text pairs in a single non-autoregressive forward pass. Unlike ASR (autoregressive, token-by-token), the forced aligner runs the entire sequence in one forward pass through the decoder. The classify head predicts a timestamp class (0–4999) at each token position, which maps to time via classindex × 80ms. This model is designed for use with speech-swift: Text decoder (attention projections, MLP, embeddings) quantized to 4-bit using group quantization (groupsize=64). Audio encoder and classify head kept as float16 for accuracy.