Tango knows when
you’re done talking

Before Tango. The caller reads out an address with short pauses. The previous agent waits for 500 ms of silence: it fires at 6.90 s just as she starts on the city, talks over her from 7.30 s and asks for the city again. She repeats it, so the agent has the whole address only at 16.58 s.

Caller

Um, my address is 6696 Lake Drive,

Caller

Milwaukee, Wisconsin…

Talked over

The silence timer fires mid-address

Previous agent

Got it. 6696 Lake Drive. What city is that in?

Caller

Milwaukee, Wisconsin, 53201.

Previous agent

Perfect. Got all of that.

  • Fast and accurate

    On LiveKit’s public end-of-turn benchmark, Tango is the most accurate model for its speed, and the fastest for its accuracy.

  • Misses fewer turn endings

    It’s the fastest of 11 systems once you require at least 90% detection, and it missed just 15 of 400 true turn endings.

  • Fewer interruptions, in production

    Tango tells a real interruption from a quick “mm-hmm”, so Simple doesn’t cut off a customer who’s simply agreeing.

  • Skips the transcript

    The old pipeline waits for speech to become text before the agent can think. Tango reads end-of-turn and interruption straight from the audio.

  • Built for real phone lines

    Tango is dual-channel and telephony-native, made for compressed 8 kHz phone audio as well as 24 kHz VoIP, on the calling systems you already run.

The details

How Tango works, in plain words.

How it listens
Tango reads the call audio directly, the caller and the agent on separate channels at once, every 80 ms, with inference in less than half that time.
Finishing
It calls the end of a turn a median 300 ms before a conventional voice-activity detector would even notice the pause, on real phone audio.
Pauses
When a caller is still thinking, Tango holds through the pause and lets them finish.
Interruptions
It recognizes a real interruption about 100 ms after the caller starts talking, then stops and listens.
Backchannels
A quick “mm-hmm” or “okay” is a nod, not an interruption, so Simple carries on.
Speed
It skips the transcript step that adds roughly a second in a cascaded pipeline. Transcripts are still made for every call, just not in the way of the reply.
Audio
Dual-channel and telephony-native, for 8 kHz phone lines as well as 24 kHz VoIP.
Benchmarks
Fastest of 11 systems on LiveKit’s public end-of-turn benchmark at 90% detection or better. With each model tuned to 5% or fewer mid-turn cut-ins, it missed 15 of 400 turn endings, against 54 for Soniox and 197 for Deepgram Flux. Against Smart Turn, it fires 3.7 to 4.7 times less often mid-speech at equal turn-end coverage.
In production
64% fewer interruptions than our previous agent. Tango is live across Simple’s production traffic.
Your systems
It plugs into the inbound and outbound calling systems you already run.

Questions

What is Tango?

Simple’s brand-new conversational awareness model. It listens to the call audio itself, not a transcript, to tell when a caller is still thinking, when they’re done, and when they’re cutting in.

How is Tango different from a speech-to-text pipeline?

A cascaded pipeline turns speech into text before it can decide anything. Tango reads end-of-turn and interruption straight from the audio, so there’s no transcript to wait for.

Will Simple talk over my customers?

Tango is built to wait its turn. It holds through pauses, carries on through an “mm-hmm”, and stops for a real interruption about 100 ms in. In production, it cut interruptions by 64%.

How do you know it’s better?

On LiveKit’s public end-of-turn benchmark, Tango is the fastest of 11 systems at 90% detection or better, and it missed 15 of 400 turn endings. In production, it cut interruptions by 64% against our previous agent.

Do I need to set anything up?

No. Tango is live across Simple’s production traffic, and it plugs into the calling systems you already run.

Never miss another call

Tell us where calls and leads slip through today. We’ll show you what Simple can catch first.