Nvidia Noise Removal is a skill that sends audio to the NVIDIA Maxine Background Noise Removal NIM over a TLS-secured gRPC connection. It accepts WAV files up to 35MB and isolates human speech from environmental sounds. The skill targets mechanical noise such as keyboard typing, mouse clicks, and PC fans, along with environmental distractions including traffic, wind, barking dogs, and HVAC hum. It preserves the natural tone and clarity of the speaker’s voice while offering an adjustable intensity ratio from 0.0 to 1.0 so an agent can trade off between noise suppression and vocal preservation. Because the skill manages the gRPC handshake and protobuf communication, an agent can submit a local audio file and receive a cleaned version without handling the underlying protocol details directly. The host environment must provide Python 3.10 or later, the grpcio package, and a free NVIDIA API key. There are no exposed tools; the agent invokes the skill itself to process the audio. This makes it suitable for automated post-processing of voice recordings where the source material contains consistent non-speech interference.

Use cases

Use it to remove keyboard clicks and mouse noise from screen-capture voiceovers before publishing.

Use it to strip HVAC hum and traffic rumble from field-interview WAV files.

Use it to clean up conference-call recordings where multiple speakers overlap with background chatter.

Use it to balance noise suppression against voice clarity by setting the intensity ratio per file.

Use it to batch-process podcast segments that suffer from fan or wind noise.

Use it to prepare noisy audio transcripts by reducing non-speech interference in the source WAV.

When to use it

Use when the input is a mono or stereo WAV file no larger than 35MB.

Use when the noise profile includes mechanical or environmental sounds rather than reverberation or echo.

Use when the host already has Python 3.10+ and a valid NVIDIA API key available.

Use when the agent should offload denoising to NVIDIA Maxine rather than run local DSP filters.

Use when adjustable intensity is needed to preserve quiet or breathy speech while still reducing background noise.

When not to use it

Do not use if the audio is not in WAV format or exceeds the 35MB input limit.

Do not use if the host cannot reach NVIDIA’s gRPC endpoint or lacks the required API key.

Do not use if the primary problem is room echo or reverberation, which the Maxine BNR model is not described to address.

Do not use if the environment cannot satisfy the Python 3.10 and grpcio dependencies.

Do not use if the agent needs discrete tools for inspection or step-by-step audio editing, because the skill exposes no tools.