The Telephony Layer Under Your Voice Agent (And When It Breaks)
Last updated on May 22, 2026
A voice agent demo sounds clean because the demo runs over a browser, in WebRTC, on a fast connection, with no carrier in the path. Production calls travel a different route. They cross a carrier network, a Session Border Controller, a codec boundary, and a caller-ID attestation chain – and every one of those stages stays invisible until it fails. The telephony layer surfaces three ways: a carrier outage drops live calls, a phone number gets labeled “Spam Likely” and answer rates collapse, or a codec mismatch forces transcoding that quietly degrades transcription accuracy. Teams that ignored this layer during the build discover it during the incident.
This article maps the telephony stack beneath a voice agent: carrier and CPaaS selection, SBC architecture, codec choice, the narrowband transcription ceiling, caller-ID regulation, failover design, and where agent platforms plug in. The audience is voice-AI engineers and technical founders deciding how much of this layer to own. One conclusion stated up front: for most voice agents the per-minute carrier cost is small next to speech-to-text, LLM, and text-to-speech cost, so owning telephony is rarely the biggest cost lever. The reasons to care about it are reliability and deliverability, not unit economics.
The telephony layer sits beneath the agent runtime
A voice agent runtime is a pipeline: voice activity detection, speech-to-text, an LLM, then text-to-speech. That pipeline produces and consumes audio. The telephony layer is everything that moves that audio between the agent and a human on a phone or in a browser.
That layer has three parts. A carrier or CPaaS provider connects to the public phone network and originates or terminates calls. A signaling and media transport protocol – SIP for phone numbers, WebRTC for browsers – carries call setup and the audio stream. A Session Border Controller polices the boundary between networks, handling security, protocol normalization, and codec conversion.
Agent platforms either bundle this layer or expose it. Turnkey platforms hide it entirely. Bring-your-own-carrier (BYOC) setups let an engineer plug a chosen carrier into a managed agent runtime. Fully in-house builds own all three parts. The rest of this article works through each part so the choice between those three models rests on evidence.
Carrier and CPaaS comparison: Twilio, Telnyx, Bandwidth, SignalWire, Plivo
The carrier is the entry point. A Communications Platform as a Service (CPaaS) provider sells programmable access to the phone network through an API, runs carrier-grade infrastructure, and bills per minute. Five providers dominate the voice-agent conversation: Twilio, Telnyx, Bandwidth, SignalWire, and Plivo.
The headline number is the per-minute rate, but the headline number misleads. A voice agent connecting through SIP pays the SIP rate, not the local-number rate, and some providers add channel or trunking fees on top. Model the full path before comparing.
The table below lists US per-minute voice rates observed on 2026-05-22. Prices change frequently – re-verify on each provider’s pricing page before committing.
| Provider | Inbound local | Inbound toll-free | Outbound US | SIP / trunking rate | Notable structure |
|---|---|---|---|---|---|
| Twilio | $0.0085/min | $0.0220/min | $0.0140/min | $0.0040/min | TwiML SIP connections tier from $0.0040/min down to $0.0007/min above 100M monthly minutes |
| Telnyx | from $0.0032/min | from $0.015/min | $0.002/min Voice API base | from $0.005/min international; toll-free outbound free | Total = API minute + SIP trunking fee + channel fees ($12/mo per channel for first 10, scaling down to $8/mo above 250) |
| Bandwidth | UNVERIFIED | UNVERIFIED | UNVERIFIED | UNVERIFIED | Tier-1 facilities-based carrier; pricing page blocks automated fetch (see note below) |
| SignalWire | $0.00660/min | $0.01470/min | $0.00800/min local | $0.00300/min (SIP and WebRTC, both directions) | Lowest flat SIP rate; Hawaii, Alaska, high-rate-center destinations add an access fee |
| Plivo | $0.0055/min | $0.0180/min | $0.0115/min local | $0.0033/min | Billed in 60-second intervals; regional add-ons for Hawaii and Alaska |
Bandwidth is a deliberate gap. The research could not verify Bandwidth’s current US rate because its pricing page returns an HTTP 403 to automated fetch. Secondary sources cite voice “starting at $0.0055/min,” but that figure is unverified and should be confirmed manually on the Bandwidth pricing page before any decision rests on it. What is verifiable: Bandwidth is a tier-1, facilities-based carrier that owns its own network and numbers, which makes it structurally different from CPaaS resellers and a common choice for enterprises that want a carrier-direct relationship.
The positioning differences matter more than the third decimal place. Twilio offers the broadest product surface and the largest ecosystem, generally carries the highest per-minute rate, and is commonly negotiated down through committed-use agreements. Telnyx owns a private global IP backbone and markets itself as a licensed carrier with a latency advantage – that latency claim is the vendor’s, not an independently measured fact, and no neutral 2026 benchmark confirms it. Telnyx also carries the cheapest headline rate but the most layered cost: API minute plus SIP trunking fee plus per-channel concurrency fees. SignalWire, built by the creators of FreeSWITCH, offers the lowest flat SIP rate at $0.003/min in both directions. Plivo sits in the mid-market with simple flat SIP pricing.
For a voice agent, the rate that applies is almost always the SIP rate, because the agent connects through a SIP trunk rather than terminating to a physical phone number it owns. That puts the relevant carrier cost in the $0.003–$0.014/min band. Hold that band in mind for the build-vs-buy section.
SBC architecture and why concurrency exposes it
A Session Border Controller (SBC) sits at the border between two VoIP or telephony networks and controls the signaling and media sessions crossing that border. SIP carries the signaling; RTP or its encrypted form SRTP carries the media. The SBC governs both.
Its core functions, per the reference description, are topology hiding, SIP normalization between dialects, NAT traversal, security against denial-of-service and fraud, access control, media transcoding, call admission control, and lawful-intercept support. SIP is not one consistent protocol in practice – every carrier speaks a slightly different dialect, and the SBC interworks between them. Some SBCs are decomposed: the signaling plane and the media plane run on separate hardware linked by a control protocol.
The capacity spec that matters is concurrent sessions, not calls per second. The defining number to size against is how many simultaneous calls the device is rated to hold. A voice agent that handles 50 calls a day at 20 concurrent peak is a different sizing problem than one handling the same daily volume at 5 concurrent. Call admission control enforces that ceiling, capping concurrent sessions to protect downstream systems from overload.
Concurrency exposes the SBC because transcoding scales with it. Each concurrent call that needs codec conversion consumes CPU on the media plane (the next section quantifies how much). Cloud-native SBCs such as Oracle’s address this by scaling media, signaling, and transcoding independently, so a transcoding-heavy workload does not force over-provisioning of the signaling plane, and SBC clusters load-balance adaptively across nodes. An SBC sized only for signaling throughput will fall over under a transcoding-heavy concurrent load.
Most voice-agent builders never touch an SBC directly. The CPaaS is the SBC. Twilio, Telnyx, SignalWire, Plivo, and Bandwidth all run carrier-grade SBCs that the agent never configures. A dedicated SBC – self-managed on hardware or as a cloud-native instance, or consumed as a managed SBC-as-a-service – becomes a real decision only when going BYOC or fully in-house. Self-managed options range from commercial appliances (Oracle Acme Packet, Ribbon, AudioCodes) to open-source software used as an SBC (Kamailio, OpenSIPS, drachtio, FreeSWITCH).
Codec choice: Opus, G.711, G.722, and the transcoding penalty
A codec encodes and decodes the audio stream. The choice sets audio quality, bandwidth, and – through transcoding – CPU cost and transcription accuracy. Three codecs cover the voice-agent case.
| Codec | Sample rate | Bitrate | Band | Role |
|---|---|---|---|---|
| G.711 | 8 kHz | 64 kbit/s fixed | Narrowband (~300 Hz–3.4 kHz) | The PSTN baseline; PCM µ-law/A-law; resilient across multiple transcoding hops |
| G.722 | 16 kHz | 48 / 56 / 64 kbit/s | Wideband (“HD voice”) | Noticeably clearer than G.711 where the full path supports it |
| Opus | 8 / 16 / 24 / 48 kHz | 6 kbit/s to 510 kbit/s | Narrowband to fullband, adaptive | The WebRTC default codec; scales from low-bitrate speech to fullband stereo |
The public phone network forces G.711. Any call that touches the PSTN is bandlimited to 8 kHz narrowband. A voice agent running Opus internally – which it will, if it uses WebRTC anywhere in its media stack – must convert Opus to G.711 and back at the SBC or gateway whenever a call bridges to the PSTN. That conversion is transcoding, and transcoding carries a measurable penalty.
The penalty has three dimensions. Capacity: real-time transcoding reportedly cuts a system’s call-handling capacity by roughly an order of magnitude. Latency and CPU: a transcoding step adds roughly 20–50 ms of latency and can consume 50–80% of a CPU core per concurrent call. Information loss: once audio crosses into G.711, the wideband detail is permanently gone. Transcoding back to Opus afterward cannot recover frequency content the narrowband hop never carried.
The practical takeaway for codec strategy: a call that stays inside WebRTC end-to-end can hold Opus and stay wideband. A call to or from a phone number will hit G.711 the moment it touches the PSTN, and no codec choice on the agent side changes that. Carriers market HD-voice or wideband SIP trunking (G.722 or Opus) to keep audio at 16 kHz across the path – but that only holds where the entire path supports it, and a single PSTN hop to a non-HD endpoint collapses the whole call back to 8 kHz.
The 8 kHz narrowband ceiling on transcription accuracy
The 8 kHz constraint is not only an audio-quality issue. It sets a hard ceiling on speech-to-text accuracy, and that ceiling is imposed by the channel, not by the model.
PSTN audio is bandlimited to roughly 300 Hz–3.4 kHz. High-frequency phonetic cues live above that range. Fricatives – the /s/, /f/, and /th/ sounds – carry energy above 3.4 kHz, and on a narrowband channel that energy is simply absent. The STT model receives audio with the distinguishing information for those phonemes removed, which raises confusability and word error rate.
The numbers come from Voicegain’s 2025 benchmark on 8 kHz call-center audio – 40 audio files across 8 customers, with accuracy reported as 1 minus word error rate:
| STT model | Accuracy on 8 kHz audio |
|---|---|
| Amazon AWS | 87.67% |
| Voicegain-Whisper-Large-V3 | 86.17% |
| Voicegain Omega | 85.09% |
| Voicegain Kappa (streaming) | <1% below Omega |
| Google Video model | 68.38% |
Best-in-class STT tops out around 86–88% accuracy on 8 kHz telephony audio. That is the ceiling. A better model moves the number a point or two; it does not break the ceiling, because the limit comes from missing acoustic information rather than from model capability. The Voicegain benchmark did not publish a matched 16 kHz comparison, so the exact narrowband-versus-wideband accuracy delta is not verified here – the verified fact is the narrowband ceiling itself.
The same constraint degrades the output side. Synthesized voices sound worse downsampled to narrowband – a TTS voice tuned for 24 kHz loses naturalness when the call carries it at 8 kHz. So the PSTN hop costs accuracy on the way in and naturalness on the way out.
This reframes prompt-and-model tuning. A team chasing the last few points of transcription accuracy by swapping STT models is optimizing above a ceiling the telephony layer already set. On a PSTN call, accuracy near 87% is close to the channel limit regardless of model. Knowing where the ceiling sits prevents wasted effort.
STIR/SHAKEN attestation and the 2025-2026 caller-ID rules
A voice agent that places outbound calls has a deliverability problem distinct from quality: the call has to be answered. Caller-ID authentication and number reputation decide whether it is.
STIR/SHAKEN is the framework US carriers use to sign calls and assert how much the originating provider knows about the caller. It defines three attestation levels:
- A, full attestation: the originating provider authenticated the customer and confirmed the customer is authorized to use the calling number.
- B, partial attestation: the provider authenticated the call origin but cannot confirm the caller owns the number.
- C, gateway attestation: the provider can verify where it received the call but not its source – typical for international or inbound-gateway calls.
For an outbound voice agent, the target is A-level attestation. Calls signed B or C are far more likely to be spam-labeled or filtered before they reach a handset. Reaching A requires the originating provider to verify number ownership, which is why number provenance – whether the carrier can confirm the agent owns the calling number – belongs in the carrier-selection decision, not just the per-minute rate.
The rules around signing tightened in 2025 and 2026. As of September 18, 2025, a provider with a STIR/SHAKEN obligation may use a third party to perform the technical act of signing only if the provider itself makes all attestation-level decisions per the technical standards and calls are signed with the provider’s own certificate from a STIR/SHAKEN Certificate Authority. The rule targets improper A and B attestations applied by parties that did not originate the call. On the compliance-filing side, the Robocall Mitigation Database annual recertification window opened February 1, 2026, with recertification due March 1, 2026.
One more rule is proposed, not adopted. The FCC’s December 2025 Notice of Proposed Rulemaking, “Advanced Methods to Target and Eliminate Robocalls,” proposes requiring terminating providers to transmit verified caller-name and caller-identity information to the handset whenever they pass an A-level attestation. Comments were due January 5, 2026 and reply comments February 3, 2026. As of 2026-05-22 the proposal is not final, and whether it has been adopted is unverified. If adopted it would raise the practical value of A-level attestation further, because the verified name would reach the called party’s screen.
A2P 10DLC and number-reputation management
Attestation handles whether a call is signed. Reputation handles whether the called party’s carrier flags the number as spam. These are separate systems, and a voice agent at volume has to manage both.
A2P 10DLC is the registration regime for application-to-person messaging over standard 10-digit long-code numbers. It is an SMS regime, not a voice regime, but it matters to voice-agent teams for two reasons: many voice agents also send SMS, and the brand identity registered for 10DLC feeds the same caller-reputation systems. Since February 1, 2025, all major US carriers block – not throttle – 100% of unregistered 10DLC traffic. Unregistered means undelivered.
Several 10DLC details changed for 2026, drawn from secondary 2026 compliance guides and worth confirming against The Campaign Registry directly before relying on them. The Campaign Registry introduced an “Authentication+” identity tier in August 2025 for publicly traded companies, carrying a $12.50 fee. A Reseller ID is mandatory when a platform, agency, or SaaS registers campaigns on behalf of a client. An EIN must be at least 15 days old to register a brand. Opt-in URLs must be live and carrier-verifiable. State-level retention rules are emerging – Virginia now requires 10-year opt-out record retention. Brand registration runs 1–3 business days and campaign registration 3–7, stretching to 10–15 in high-volume periods, so registration belongs early in a launch timeline.
Number reputation is the voice-side equivalent. Three analytics engines drive carrier spam-labeling: Hiya on AT&T, TNS on Verizon, and First Orion on T-Mobile. A high-volume outbound agent with low answer rates will accumulate spam flags over time, and once a number reads “Spam Likely” on a handset, answer rates collapse.
The remediation path is documented. Free Caller Registry is a no-cost centralized portal that submits number registrations to all three engines and is the standard first step to register legitimate business numbers and remediate a “Spam Likely” label. Paid managed services exist for branded caller ID and ongoing monitoring – Numeracle and Bandwidth both offer number-reputation management products. For a high-volume outbound agent the working practice is to rotate a healthy pool of numbers, register them on Free Caller Registry, hold A-level attestation, and monitor for flags – because labeling is reputation-based and degrades with high volume and low answer rates.
WebRTC versus SIP: matching protocol to deployment
SIP and WebRTC are the two transport protocols beneath a voice agent, and the choice is not a quality ranking. It follows from where the human is.
SIP is the protocol for phone-number-based calls across the PSTN and mobile networks. An outbound AI SDR dialing cell phones must use SIP at the carrier edge – there is no other path to a phone number. SIP carries signaling; RTP or SRTP carries the media.
WebRTC is the protocol for browser and in-app voice. It uses UDP-based media transport, Opus by default, built-in DTLS-SRTP encryption, and ICE/STUN/TURN for NAT traversal. An in-product “talk to the agent” widget runs on WebRTC.
The decision logic is direct. Calls to or from phone numbers route over SIP. Browser or app voice routes over WebRTC. When the latency target is aggressive – sub-300 ms, with transport as the binding constraint – WebRTC’s UDP transport wins. A deployment inside an existing SIP contact center makes SIP integration the path of least resistance; a greenfield deployment gets a cleaner architecture starting from WebRTC.
Most production voice agents run both. WebRTC carries audio inside the media stack and SIP connects at the carrier edge, which means a call crosses a SIP-to-WebRTC bridge – and a bridge means real-time transcoding and codec negotiation, with the 20–50 ms per step and 50–80%-of-a-core costs from the codec section. LiveKit ships such a bridge as a documented component. The protocol choice is not about which is better; it is about matching protocol to deployment context, and accepting the bridge cost where both appear.
Multi-carrier failover under outage
A single carrier is a single point of failure. A 2020 US carrier outage left tens of millions without phone service for roughly 15 hours – a voice agent wired to one carrier is exposed to exactly that.
The multi-carrier pattern qualifies each phone number across more than one carrier network. If carrier A fails, traffic for that number reroutes to carrier B, ideally before the caller notices. A failover design combines several building blocks: independent trunks on independent carriers, SBCs performing health-checked routing and load balancing, SIP OPTIONS probes as the keep-alive and health-check mechanism, DNS and IP-level routing with PSTN fallback, per-route priorities with documented disaster-recovery routing plans, and multi-region cloud deployment.
Detection speed depends on configuration. A well-configured environment detects failure within 5–30 seconds, set by the SIP OPTIONS probe interval and retry thresholds, and sub-second switchover is achievable so users rarely notice. A widely cited figure puts telecom downtime at roughly $5,600 per minute – treat that as unverified, since the original source is not traced, and use it as illustration rather than a planning input.
For a voice agent, failover has to cover both the carrier layer and the AI pipeline. Number-level multi-carrier routing protects inbound traffic. An outbound agent should hold credentials for at least two trunks and route per health-check status. One counter-argument deserves a mention as a tradeoff rather than consensus: some single-provider vendors argue that redundancy within one provider’s network simplifies operations and reduces the integration surface. That is a real operational simplification, but it leaves the provider itself as the single failure domain.
How agent platforms sit on the telephony layer
The telephony layer – carrier, SIP or WebRTC, SBC – sits beneath the agent runtime of VAD, STT, LLM, and TTS. Agent platforms differ in how much of that layer they expose, and the difference maps onto the build-vs-buy decision.
- Vapi is turnkey. It abstracts transport entirely, moving audio over WebSockets with the Twilio and SIP connections fully hidden. An engineer never configures transport directly. Vapi supports BYOC.
- Retell is a managed cloud service that supports BYOC through custom SIP trunks, with community integrations documenting connection via jambonz.
- Bland AI is a turnkey voice-agent platform that documents SIP integration as an enterprise feature with full TLS and SRTP encryption.
- LiveKit is an open-source SFU where the agent joins a Room as a headless participant. It ships a dedicated SIP-to-WebRTC bridge and supports explicit BYOC: an inbound trunk restricts incoming calls to a chosen provider, and an outbound trunk holds provider credentials and endpoint, with Telnyx, Twilio, and Plivo named as providers.
- Pipecat is a transport-agnostic Python framework from Daily. The engineer chooses the transport and the pipeline logic stays unchanged – maximum BYOC flexibility, most assembly required.
- Telnyx AI layers voice-agent capability directly on the Telnyx carrier network, delivering telephony and AI from one vendor. It is the opposite of BYOC: everything sits in-house at the vendor.
The platforms occupy a spectrum. Vapi and Bland hide the telephony layer; LiveKit and Pipecat expose it; Telnyx AI owns it end-to-end. Where a team sits on that spectrum should follow from how much of the telephony layer it actually needs to control, which is the next question.
Decision framework: turnkey, BYOC, or in-house
The honest starting point: for a voice agent, carrier per-minute cost is small next to the rest of the pipeline. The carrier SIP rate sits in the $0.003–$0.014/min band. Platform all-in costs run far higher – secondary-source estimates put Vapi at $0.13–$0.33/min all-in (including a ~$0.05/min platform fee) and Retell at $0.13–$0.31/min. The dominant costs are LLM and TTS. Carrier minutes are a rounding error against them. So the decision to own telephony should be driven by control and reliability, not by chasing per-minute savings.
Turnkey platform. A turnkey platform fits when telephony is not a differentiator and speed to launch is. The platform runs carrier-grade SBCs, handles attestation, and abstracts transport. The tradeoff is less control over routing, codec path, and failover, and a per-minute markup. For most early-stage voice agents, that markup buys time that is worth more than the margin.
BYOC. Bringing a chosen carrier into a managed agent platform fits when the carrier relationship needs to be specific – a particular provider for number provenance and A-level attestation, a negotiated committed-use rate, an existing contact-center trunk, or a multi-carrier failover requirement the platform’s bundled carrier cannot meet. BYOC captures most of the carrier-side control without rebuilding the media stack. LiveKit, Vapi, Retell, and Pipecat all support it.
Fully in-house. Owning carrier contracts, SBCs, and routing fits a narrow case. Generic CPaaS analysis puts the crossover above roughly 50M monthly minutes and with 3+ dedicated telecom engineers, estimates the initial build at $250K–$500K over a 9–12-month timeline against under 4 weeks for a pre-built CPaaS, and flags vendor API margin above ~$50K/month as the trigger to audit carrier-direct pricing. Treat those numbers as directional and as understatements of the relevant gap – they come from messaging and CPaaS economics, not voice-agent economics, and no voice-agent-specific crossover analysis was located. In-house buys margin elimination, routing and codec control, and no per-minute lock-in, at the cost of carrier contracts, STIR/SHAKEN signing obligations, Robocall Mitigation Database filings, E911, number provenance, 24/7 operations, and SBC maintenance.
The framework in one line: default to turnkey, move to BYOC when the carrier relationship has to be specific for deliverability or contract reasons, and go fully in-house only at sustained scale with a dedicated telecom team. The reasons that actually move a voice agent off turnkey are A-level attestation, number provenance, and multi-carrier failover – reliability and deliverability concerns – not the per-minute rate.
Softcery builds production voice agents for B2B SaaS founders, and the recurring pattern is that the telephony layer gets attention only after an incident. Choosing carrier, codec path, and failover model deliberately during the build is cheaper than discovering them during an outage or a spam-label investigation.
Frequently Asked Questions
Frequently Asked Questions
Rarely. The carrier SIP rate for a voice agent sits in the $0.003–$0.014/min band (prices observed 2026-05-22 and subject to change), while all-in platform costs run an order of magnitude higher because LLM and TTS dominate the pipeline. Secondary-source estimates put Vapi at $0.13–$0.33/min and Retell at $0.13–$0.31/min all-in. Owning telephony eliminates a small fraction of total cost. The real reasons to take more control of the telephony layer are A-level caller-ID attestation, number provenance, and multi-carrier failover, which are reliability and deliverability concerns rather than cost concerns.
Calls that cross the public phone network are bandlimited to 8 kHz narrowband, roughly 300 Hz to 3.4 kHz. High-frequency phonetic cues for fricatives like /s/, /f/, and /th/ live above 3.4 kHz and are absent from the audio entirely. Voicegain’s 2025 benchmark on 8 kHz call-center audio showed best-in-class speech-to-text models topping out at 86–88% accuracy, with Amazon AWS at 87.67%. That ceiling is set by the channel, not the model, so swapping STT models moves the number a point or two but does not break the ceiling. A call that stays inside WebRTC end-to-end avoids the narrowband hop and is not subject to this limit.
Transcoding is converting media from one codec to another mid-call, most commonly Opus to G.711 when a WebRTC-based agent bridges to the PSTN. Each transcoding step adds roughly 20–50 ms of latency, can consume 50–80% of a CPU core per concurrent call, and reportedly cuts a system’s call-handling capacity by roughly an order of magnitude. It also causes permanent information loss: once audio crosses into 8 kHz G.711, the wideband detail cannot be recovered by transcoding back. A voice agent pays this penalty on any call that touches a phone number, because the PSTN forces G.711.
A-level, full attestation. Under STIR/SHAKEN, A-level means the originating provider authenticated the customer and confirmed the customer is authorized to use the calling number. Calls signed B (partial) or C (gateway) are far more likely to be spam-labeled or filtered before reaching a handset. Reaching A requires the carrier to verify number ownership, which makes number provenance a real carrier-selection criterion. As of September 18, 2025, FCC rules also restrict third-party call signing: the obligated provider must make all attestation decisions and sign with its own certificate.
Bring-your-own-carrier fits when the carrier relationship has to be specific. Common triggers: needing a particular provider for number provenance and A-level attestation, a negotiated committed-use rate, an existing SIP contact-center trunk to integrate with, or a multi-carrier failover design the platform’s bundled carrier cannot support. BYOC captures carrier-side control without rebuilding the media stack, and LiveKit, Vapi, Retell, and Pipecat all support it. A fully in-house build is a separate, higher bar – generic CPaaS analysis places the crossover above roughly 50M monthly minutes with a dedicated telecom team, and no voice-agent-specific crossover figure is established.
See how much it would cost to build and launch your AI voice agent — tailored to your business in under a minute.
Try the AI Voice Calculator
