DialogueSidon

Two-speaker dialogue separation and restoration, with 24 kHz mono stems. Converted from sarulab-speech/DialogueSidon for use with mlx-audio.

from mlx_audio.sts import load
from mlx_audio.audio_io import write

model = load("mlx-community/DialogueSidon-bf16")
result = model.separate("dialogue.wav", num_steps=30, seed=0)
for i, speaker in enumerate(result.speakers, 1):
    write(f"speaker_{i}.wav", speaker, result.sample_rate)

The two channels are anonymous speaker slots. Long recordings use overlapping chunks and waveform correlation to maintain channel assignment. Consistency across silence is not guaranteed. This is an offline generative model, so the stems need not sum to the mixture. The default chunk and overlap are 20 and 5 seconds; use chunk_seconds=None for whole-file inference when memory permits.

The checkpoint retains CC-BY-NC-4.0. Original model by Wataru Nakata, Yuki Saito, Kazuki Yamauchi, Emiru Tsunoo, and Hiroshi Saruwatari (SaruLab). See the original model card and Sidon implementation.

Downloads last month: 41

Safetensors

Model size

0.4B params

Tensor type

BF16

MLX

Hardware compatibility

Quantized

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for mlx-community/DialogueSidon-bf16

Base model

sarulab-speech/DialogueSidon

Finetuned

(2)

this model