FlashSR
此 Soniqo 页面记录本地 speech-swift / speech-core 实现中的 FlashSR。Hugging Face 包链接放在集成说明之后。
先进入站内页面
首页卡片和文档菜单先指向这里；源模型和权重包链接仍在本页提供。
概览
| 模型 | FlashSR |
|---|---|
| 用途 | Audio super-resolution for low-bandwidth or lossy audio |
| 后端 | MLX int4 default; int8 available |
| 输出 | 48 kHz mono waveform, same length as input |
| 语言 | Audio-content agnostic |
| 许可证 | MIT |
| 状态 | Ready through speech upsample and the FlashSR Swift product |
| 来源 | FlashSR / AudioSR distillation |
| Swift 产品 | FlashSR |
| CLI / 运行时 | speech upsample |
使用
下面的片段对应当前 speech-swift 仓库暴露的 API 或命令。
# Upsample a low-bandwidth recording to 48 kHz mono.
.build/release/speech upsample noisy_lowres.wav \
--variant int4 \
-o clean_hr.wav模型链接
实现说明
- Download requests the single model.safetensors bundle and config explicitly; this is simple enough that byte-weighting is less critical than for sharded models.
- Input is resampled to 48 kHz mono and processed in non-overlapping 5.12 second windows.
- INT4 is a download-size optimization; runtime weights dequantize to FP, so memory footprint matches int8.
- The model conforms to SpeechEnhancementModel, but its semantics are super-resolution rather than denoising.