Upscale Audio (MossFormer2)

How it fits

Typical backends

Caveats

• I/O contractOutput is ALWAYS mono at 48 kHz regardless of input rate or channel count; multi-channel input is mixed to mono without weighting. The model operates entirely at 48 kHz, so a low-rate input is resampled UP into a 48 kHz container before inference — it is not downsampled.
• I/O contractEvery input AudioFrame yields exactly one output AudioFrame. An empty input frame is returned unchanged at its original rate without invoking the model, so it is the one case the output is not 48 kHz. A clip too short for the model's STFT/band-split front end is passed through resampled to 48 kHz, without bandwidth extension.
• LimitationMossFormer2_SR_48K is trained on clean speech: music and general or noisy audio come back with little meaningful gain. Run a speech denoiser such as denoise_audio_mossformer2 first if the input is noisy.
• LatencyEach input is super-resolved independently with no cross-call buffer. The ClearVoice runtime decodes clips at or under its ~20 s threshold in a single window and sliding-window decodes longer clips, so there is no external padding. The output sample count tracks the 48 kHz input length, not the original input sample count.
• State lifetimedevice is fixed at startup with no runtime mutation; changing it requires a redeploy, and it falls back to CPU when no GPU is present — CPU mode is well below real-time for this model. The MossFormer2_SR_48K weights are fetched by the ClearVoice runtime on first deploy and cached, so the first call also pays the download and cold-kernel cost.
• Compatibilitydenoise_audio_mossformer2 is the denoising sibling on the same ClearVoice runtime; `upscale_audio_audiosr` and `upscale_audio_universr` are alternative super-resolution models.