What are the top voice evaluation platforms for speech agents?

As of October 2026 the four dedicated platforms are Hamming, Roark, Coval and Cekura. Hamming is the enterprise pick for large-scale simulation, red teaming and monitoring (50,000+ concurrent test calls, 65+ languages; pricing by quote). Roark is the cheapest way to start with usage-based pricing: $50 free credit, then $0.15 per simulated minute plus provider costs, or $0.10 on the $500-a-month Team plan. Coval has the clearest fixed plans ($100 Starter with 100 simulation minutes, $500 Growth with 1,000). Cekura is best for healthcare teams on a budget: $0.25 per test minute pay-as-you-go and a signed BAA on its $500 Startup plan.

How much does voice agent testing cost?

List prices in October 2026: Roark charges $0.15 per simulated minute plus telephony and speech costs on pay-as-you-go and $0.04 per metric per minute for scoring; Cekura charges $0.25 per voice test minute and $0.05 per monitored call; Coval includes 100 simulation minutes in its $100 Starter plan with $0.40 per extra minute, and 1,000 minutes in its $500 Growth plan with $0.25 per extra minute. Hamming does not publish prices. A team running 2,000 simulated minutes a month pays roughly $300 to $500 on any of the published plans.

What should a voice agent evaluation platform test?

Five things: scripted and auto-generated scenarios run as real phone or WebRTC calls; acoustic conditions such as accents, background noise and interruptions; tool calls and workflow order (did the agent actually book the appointment it promised); compliance and red-team cases such as PHI handling and social engineering; and production monitoring that scores live calls with the same metrics so regressions surface after launch.

Do I need a separate eval tool if my voice platform has analytics?

Not at small scale. Retell, Vapi and similar platforms log calls and transcripts, which is enough for one agent and low call volume. Add a dedicated evaluation platform when you ship prompt or model changes weekly, run several agents, operate in a regulated industry, or need load tests before launch, because only the dedicated tools replay hundreds of simulated callers against each new version and gate releases in CI.

Quick Answer

Top Voice Evaluation Platforms for Speech Agents (2026)

Published: October 8, 2026

The short answer

Pick Hamming if you need enterprise-scale simulation and red teaming, Roark if you want usage-based pricing from $0, Coval if you want fixed monthly plans, and Cekura if you are a healthcare team that needs a BAA cheaply. All four simulate callers against your voice agent, score the calls with automated metrics and monitor production traffic; they differ in scale, pricing model and compliance packaging. Prices below are list USD from each vendor’s pricing page, read October 8, 2026.

The comparison

Platform	Best for	Entry price	Simulation rate	Monitoring	Compliance
Hamming	Enterprise QA: testing, red teaming, monitoring	Quote only	Not published	Live-call monitoring for drift and broken promises	SOC 2 Type II
Roark	Usage-based start, many integrations	$0 + $50 credit; Team $500/mo spent as usage	$0.15/min + provider costs (Team $0.10, Enterprise from $0.05)	Metrics at $0.04/metric/min ($0.02 Team)	SOC 2 Type II, HIPAA BAA
Coval	Predictable fixed plans	Starter $100/mo	100 min included, $0.40/min overage (Growth: 1,000 min, $0.25)	1,000 monitored calls (Growth 10,000)	SOC 2 Type II, HIPAA, GDPR on every plan
Cekura	Healthcare and budget teams	Pay-as-you-go, first user free	$0.25 per voice test minute	$0.05 per monitored call	HIPAA, GDPR; signed BAA on $500 Startup

Platform notes

Hamming positions itself as one platform for three jobs: pre-launch testing, red teaming and production monitoring, for both voice and chat agents. It advertises 50,000+ concurrent test calls, 65+ languages and accents, and 95–96% agreement with human evaluators (vendor figures). Scenarios are generated from your prompt and scored by 50+ built-in metrics, with accents, interruptions, background noise, DTMF and IVR trees, and a CI/CD hook so a failing build does not ship. Customers named on its site include Bland Labs and Grove AI. Pricing is by quote, so it suits teams with a procurement process rather than a credit card.

Roark is usage-based in plain dollars, with no seat or per-metric licence. Pay-as-you-go starts with $50 of free credit and charges $0.15 per simulated minute plus pass-through telephony and speech costs. The $500 Team plan is spend, not a fee: it lowers simulation to $0.10 a minute and metrics to $0.02 per metric-minute and includes 25 concurrent lines and 10 seats. Scoring live production calls costs only the metric rate, because Roark did not place the call. It simulates over real PSTN and WebRTC in 45 languages and integrates with Vapi, Retell, LiveKit and Pipecat.

Coval sells fixed tiers: Starter at $100 a month (100 simulation minutes, 1,000 monitored calls, 30-day retention, 10 metrics per call, 5 concurrent simulations) and Growth at $500 (1,000 minutes, 10,000 calls, 90-day retention, unlimited metrics, 25 concurrent simulations). Enterprise starts at $4,500 a month with SAML SSO, VPC and data residency. Unlimited seats on every plan and a 20% annual discount make it the easiest to budget. A 7-day trial requires a card.

Cekura charges $0.25 per voice test minute and $0.05 per monitored call with 10 concurrent calls on pay-as-you-go, plus 300 free credits (about 60 test minutes). The $500 Startup plan covers about 2,000 test minutes and 10,000 monitored calls, 50 concurrent calls, and a signed BAA and DPA — the cheapest published route to a BAA among the four. It markets itself heavily at healthcare voice agents (workflow order, tool calls, PHI handling).

How to choose

Regulated enterprise, high call volume, needs red teaming: Hamming. Ask for load-test limits and the metric list in the quote.
Startup shipping weekly on Vapi, Retell or LiveKit: Roark pay-as-you-go, then Team once usage passes about $500 a month (about 3,300 simulated minutes at $0.15).
Finance wants a fixed number: Coval Starter or Growth.
Healthcare with a small budget: Cekura Startup for the signed BAA.

A rule that holds on every platform: simulated minutes are the cost driver, production scoring is cheap. Generate scenarios once, run the full suite on every release, and score 100% of production calls with a few high-signal metrics (task completed, tool call correct, policy followed) rather than dozens.

What changed since spring 2026

Our April comparison of Roark vs Coval vs Hamming predates two shifts: Cekura now publishes per-minute prices that undercut most plans, and Roark moved to dollar-denominated usage pricing with a $50 free start. For the agents being tested, see the best AI voice agent platforms.

Last verified: October 8, 2026. Prices are list USD from vendor pricing pages and exclude telephony and speech-model costs unless stated.

The short answer

The comparison

Platform notes

How to choose

What changed since spring 2026

Sources