Everything we found out, in the order we found it.
We are asking an industry to accept our measurements, so the reasoning has to be visible. These are the working notes: what the published claims say, where they contradict each other, what our own harness measured, and what it still cannot tell you.
- 24 Oct 20267 min
ai-coustics alternatives for voice agents and call centers (2026)
Why teams look past ai-coustics Quail Voice Focus, and a sourced shortlist as listed in October 2026: Krisp VIVA, AssemblyAI Voice Focus, LiveKit Cloud voice isolation, NVIDIA Speaker Focus, Anecho and open-source baselines. We have benchmarked ai-coustics against our own model in live conversations: parity with faster-whisper, a v8_2 lead with Gemini Live, fewer false barge-ins for ai-coustics, and the condition decides who leads.
Anecho Engineeringcomparisonai-cousticsspeaker-isolationalternatives - 22 Oct 20267 min
Krisp alternatives for voice agents and call centers (2026)
Why teams look past Krisp VIVA, and a sourced shortlist as listed in October 2026: ai-coustics Quail Voice Focus, AssemblyAI Voice Focus, LiveKit Cloud voice isolation, NVIDIA Speaker Focus, Anecho and open-source baselines, with where each runs, its public price and whether we have measured it. We have not measured Krisp.
Anecho Engineeringcomparisonkrispspeaker-isolationalternatives - 20 Oct 202612 min
Speaker isolation SDKs compared (2026): Krisp, ai-coustics, AssemblyAI, LiveKit and Anecho
A sourced comparison of speaker isolation for voice agents and call centers as listed in October 2026: where each product runs, declared latency, sample rates, integration, public price and licensing. Krisp VIVA and Anecho declare 15 ms on CPU; ai-coustics lists under 30 ms; AssemblyAI and LiveKit sell isolation inside a cloud service. Quality numbers only from our own benchmark.
Anecho Engineeringcomparisonspeaker-isolationkrispai-coustics - 20 Oct 202612 min
Live conversations: v8_2 vs ai-coustics Quail VF 2.2 S under two recognisers, and the condition decides who leads
200 constructed 30–60 s conversations, real voices and measured rooms mixed by us, streamed whole through raw audio, v8_2 and ai-coustics Quail VF 2.2 S and scored by two recognisers. With faster-whisper base.en it is parity and no engine beats raw (37.82%, 38.46%, 38.59%). With Gemini Live input transcription v8_2 beats raw and leads VF S pooled (40.50%, 36.40%, 42.87%). VF S leads with a distant talker at −10 dB, holds the pauses quieter and triggers fewer false barge-ins.
Anecho Engineeringbenchmarkspeaker-isolationconversationresults - 17 Oct 20264 min
The state of voice-AI audio, October 2026
A snapshot as of 2026-10-17: Null Test scores eleven enhancers against a do-nothing control across nineteen conditions, the control sits at 14.32% WER and our v8_2 row at 16.59%, and the market moved towards primary-speaker isolation built into platforms and STT.
Anecho Engineeringbenchmarkmarketnulltestresults - 16 Oct 20263 min
Voice isolation pricing in 2026: public per-minute rates compared
Public per-minute voice-isolation rates in October 2026 run from $0.0010 (ours, after 10,000 free minutes) and $0.0012 (LiveKit) to $0.005 per direction (Telnyx); two vendors do not publish a price. Every figure below carries its source and retrieval date.
Anecho Engineeringpricinglivekitcomparisonvoice-isolation - 15 Oct 20263 min
How do you add speaker isolation to a LiveKit Agents worker?
Run the Anecho SDK on the inbound audio inside the worker's stt_node, in 240-sample hops at 16 kHz. There is no LiveKit plugin package, so this post shows the code you write yourself, with resampling notes and the numbers.
Anecho Engineeringlivekitintegrationsdkvoice-agents - 15 Oct 20265 min
Our row is on the table: we scored our own engine on our own benchmark, and it did not beat the control either
Null Test now scores eleven enhancers against a do-nothing control across nineteen conditions, including a two-voice condition for every engine. The pooled control sits at 14.32% WER. Our v8_2 row sits at 16.59%. Nothing on the table beat the control, our engine included, and we published the row anyway — because a vendor benchmark is only worth reading if the vendor's own row can lose on it.
Anecho Engineeringnulltestbenchmarkfocusresults - 15 Oct 20265 min
How to evaluate speaker isolation: a six-question checklist for any engine, including ours
Target-referenced WER on a corpus you can download, insertions counted separately, a do-nothing control that is allowed to win, latency stated as algorithmic delay plus block budget on one core, a straight answer on where the audio goes, and a per-minute price you can put beside the ecosystem's public anchor. Six questions, each with a pass and a fail you can execute before the pilot.
Anecho Engineeringevaluationspeaker-isolationbenchmarkwer - 13 Oct 20263 min
How much does background noise cost a contact center?
Under the stated assumptions, background noise costs a 500,000-minute-a-month contact center roughly $474 to $6,000 a month. The WER inputs are measured; the minutes, repeat-rate and cost-per-minute inputs are assumptions that you should replace with your own.
Anecho Engineeringcall-centerwercostbenchmark - 09 Oct 20263 min
Why does processing audio in-process matter for GDPR and voice AI?
Processing audio in-process keeps the recording inside your own system: no extra copy, no new processor or sub-processor to contract, and no new place the audio is stored or transferred. This is engineering context, not legal advice.
Anecho Engineeringgdprprivacysdkvoice-agents - 06 Oct 20263 min
How do you use Anecho speaker isolation as a Pipecat audio filter?
Write a small Pipecat BaseAudioFilter that re-chunks the transport's inbound audio into 240-sample hops at 16 kHz and runs the Anecho SDK on each hop. It is code you write, not a published plugin; the full filter and its numbers are below.
Anecho Engineeringpipecatintegrationsdkvoice-agents - 02 Oct 20264 min
Why turn detection breaks on TV audio
A television is a second speaker, and a voice-activity detector flags speech-like frames whatever voice they belong to. In our matrix Silero VAD marks 58.2% of non-target frames as speech when a competing voice sits at 0 dB, against 7.0% on clean audio.
Anecho Engineeringvadturn-takingbenchmarkbarge-in - 29 Sept 20263 min
Why do voice agents fail in noisy calls? Five failure modes, with numbers
Voice agents fail in noisy calls because the recogniser, the voice activity detector and the turn logic all treat every voice and every codec artefact as the caller. We list five failure modes and tie each to a number from our Null Test run.
Anecho Engineeringvoice-agentsbarge-inbenchmarkstt - 25 Sept 20263 min
8 kHz vs 16 kHz for voice agents: what the WER data says
On our narrowband test an 8 kHz leg costs a recogniser 5.79% WER on clean speech against 4.74% at 16 kHz, and the gap widens as noise rises. No enhancement path we measured, native 8 kHz or upsampled, beat the untouched 8 kHz audio.
Anecho Engineeringtelephonynarrowbandbenchmarkstt - 22 Sept 20262 min
What is speaker isolation, and how is it different from noise cancellation?
Speaker isolation keeps one speaker's voice and removes everything else, including other people talking. Noise cancellation removes non-speech noise and keeps all speech. Here is the definition, a comparison, and what our own benchmark says about the two-voice case.
Anecho Engineeringspeaker-isolationexplainerbenchmarknoise-cancellation - 16 Aug 202617 min
Amazon Connect, Twilio, NICE CXone and Avaya: four media paths, two places to stand
We read the published specifications for the real-time audio surfaces of four contact-centre platforms and asked one question of each: can a third party receive the audio, process it, and have the processed audio be what the platform's own downstream consumers hear? Twilio and Avaya document yes. Amazon Connect and NICE CXone document no. Along the way: Amazon publishes 8 kHz and 'raw PCM' and nothing else, Avaya is the only one of the five documenting a wideband codec, and Twilio never says the thing we have been saying they say.
Anecho Engineeringtwilioamazon-connectnice-cxoneavaya - 16 Aug 20269 min
The resampler that ate 92% of a microphone
soxr.ResampleStream is a burst emitter. Fed the browser's 128-sample AudioWorklet quantum it returns an empty array on 92% of calls and then hands back 95 ms at once. We read that emptiness as 'still priming' and skipped the block — which also skipped the untouched microphone buffer sitting next to it, so 8% of the caller ever reached the model. Our own telemetry reported it as 0.08 for days. This is a control-flow bug, not the resample-tax claim we retracted, and here is the four-line probe that finds it.
Anecho Engineeringdspresamplingvoice-agentsdebugging - 16 Aug 202612 min
Genesys AudioHook is 8 kHz µ-law, and there is nowhere to stand
The Genesys Cloud AudioHook protocol pins its sample rate in the type system: MediaRate is the literal 8000, and 16000 appears nowhere in the specification. AudioHook Monitor is a one-way tap whose server output is explicitly discarded; Audio Connector can play audio back, but it is a bot fork inside the IVR that pauses the flow and never reaches an agent. A spec-by-spec reading of what a third party can and cannot do in that media path, with the four things we could not confirm named as unconfirmed.
Anecho Engineeringgenesystelephonycontact-centrenarrowband - 16 Aug 202610 min
audio_stream_end=True does not end a turn in the Gemini Live API
Measured on Vertex against gemini-live-2.5-flash-native-audio with one 5-second clip: audio alone returns 0 bytes, audio plus audio_stream_end=True returns 0 bytes, and audio plus 1.5 seconds of trailing silence returns 229,994 bytes. The server ends a turn when its own VAD hears silence, so your uplink has to carry a pause as real samples and never as absent chunks. Google's own documentation says two different things about this field, and the Vertex docs do not mention it at all.
Anecho Engineeringgemini-livevoice-agentsvadturn-taking - 13 Aug 202614 min
Null Test v0.1: ten speech enhancers, eighteen conditions, and nothing beat raw on the pooled average
The harness, the dataset, the configs and now the results. Ten enhancement engines scored against a raw control on WER, error decomposition, VAD and cost. Raw won on pooled WER — but at least one engine beat raw in 11 of the 18 individual conditions, and insertions went up rather than down.
Anecho Engineeringnulltestbenchmarkmethodologyopen-source - 09 Dec 202512 min
The split pipeline: why your VAD and your STT want different audio
Turn detection wants a clean stream. Transcription wants the raw one. We now have our own numbers for both halves of that claim: enhancement cost us five points of pooled WER and cut VAD false alarms by a third — while beating raw outright in 11 of 18 individual conditions.
Anecho Engineeringarchitecturevadturn-takingsdk - 18 Nov 202511 min
Does noise suppression actually help speech-to-text? We measured it.
Krisp markets a 46% WER reduction. We ran ten enhancement engines across eighteen conditions against a raw control. None beat doing nothing on pooled word error rate, though at least one engine did beat raw in 11 of the 18 individual conditions — and insertions went up rather than down.
Anecho Engineeringspeech-to-textnoise-suppressionbenchmarkswer