Questions we hear.

Voice isolation for Indian telephony. A library that removes background voices and noise from a caller's audio, 20 ms at a time, on your own servers, before voice activity detection and speech recognition hear it.

It removes what noise cancellation removes, such as traffic, fans and a busy street, and also a second person talking near the caller. Speech recognition then hears only the caller.

Yes. A second voice close to the caller or across the room is removed, and the caller is kept. Our published result is on 1,152 Hindi test calls at 8 kHz, each with a second voice and noise.

It is a library you run on your own servers, with packages for Python, Go, Rust and C, and a FreeSWITCH module. Access is private: request access, and the trial licence comes with the evaluation kit.

Yes. packethush is built for 8 kHz telephone calls and measured on them. 16 kHz audio works through the same library.

It works on the audio before any speech recognition. Our published results are on Hindi calls, and more Indian languages are being measured. Ask us about yours.

It works on sound, not words, so code-mixing needs nothing special. Our Hindi test calls are full of English words, as real calls are.

Any. It hands clean audio to whatever comes next. The gain depends on the engine, so the evaluation kit measures it with yours.

Open one session per call leg and pass it every packet. Packages for Python, Go, Rust and C, and a FreeSWITCH module.

PCM16, float, μ-law and A-law, in chunks of any size. Output has the same format and length as the input.

It adds 12 ms on top of each 20 ms packet.

No. It runs on CPU: 280 simultaneous calls on one 8-vCPU server.

No. Only licence counters are sent. The security page lists every field.

Audio passes through unchanged. A call is never dropped or muted.

Yes. The trial licence comes with the evaluation kit, so you decide with your own numbers.

By proposal, in rupees, after the trial. Ask us when we talk.