Voice FAQs

Bird Voice connects phone systems and applications to the telephone network. Use SIP trunks, browser calling and call sequences, or connect an AI voice runtime. Follow calls through their records, sessions and events.

Create a workspace, prepare a call-ready trunk and caller identity, and enable the destination you want to test. Setup time depends on access, verification and the client you connect. The first-call guide walks through a controlled browser call.

Three things on your side: a SIP trunk with your equipment allowed on it (by IP range or API key), a verified caller ID (the number you present as the caller), and the destination country enabled. Bird attaches routing to your workspace, which is the fourth prerequisite and happens on Bird's side.

No. Use the browser phone or a compatible SIP softphone, PBX or application. Match the supported transport, authentication and media settings to the trunk configuration.

Yes. You can place calls from the dashboard Phone app using WebRTC.

Per call, at a rate that depends on the destination country. Each rate carries a billing increment: a minimum billed time, then the step it rounds up to after that. A rate with a one-minute minimum and six-second steps charges a 10-second call for a full minute.

Billable time runs from the moment the called number picks up to the moment the call ends. Review the published rate components for your use case; number rental and other configured services are separate from answered call time.

Prepaid, from your organization's wallet. Bird checks your balance before dialing the carrier, so a call your wallet cannot cover is refused up front with insufficient_balance rather than billed afterwards.

Yes. A daily voice spend ceiling applies per organization, resetting at the start of each UTC day. Past it, calls are refused with daily_spend_exceeded. The amount depends on your plan, and Bird can raise it on request.

Open the call in the call log. Cost appears on the record once it has been rated, at full precision, net of tax. For totals across calls, export the filtered call list as CSV from the Calls page, or check your invoices and usage.

Three ceilings: how many calls you may have in progress at once (concurrent calls), how many new calls you may start per second (calls per second), and how much you may spend on voice in a UTC day (daily spend). The applicable caps depend on your account configuration.

The call is refused at setup, before a carrier is ever dialed. The call record names which limit was hit: concurrent_calls_exceeded, calls_per_second_exceeded, or daily_spend_exceeded. Your phone system sees a SIP 503.

Contact support with your expected traffic, destinations and spend to request a limit review. Confirm the approved limits before increasing production traffic.

Check which reason the call record carries. A dialer can hit calls_per_second_exceeded while nowhere near the concurrent-call ceiling, because the two limits are independent. Slow the dial rate and retry; retrying immediately gets the same answer.

No. SIP trunking connects a compatible phone system to Bird. Match the trunk’s supported transport, authentication and media settings to the client you use.

Yes. Configure each direction separately. Incoming calls also need an inbound-capable number pointed at the trunk and a delivery gateway.

Yes, when its runtime can handle both the SIP signaling and audio connection. The conversational application controls its speech, model and business actions; Bird carries the telephone connection.

Numbers pointing at that trunk return to a rejecting route. Re-enabling the trunk does not restore their previous bindings; review the numbers before resuming traffic.

Bird Voice does not carry emergency calls. Use a phone service designed for that requirement.

A SIP trunk is the connection between your phone system and Bird. SIP (Session Initiation Protocol) is the language phone systems speak to set up calls, and a trunk is the line those calls travel over. Bird issues your workspace a trunk address, and you point your phone system at it.

Most workspaces need only one. Create more when you want separate access rules per site or per system, since the IP allow list, the allowed API keys, and the Digest settings are all per trunk.

The trunk's SIP domain (copied from the trunk page), the username bird, and the password (the secret of an API key allowed on the trunk). Send calls to the SIP domain on port 5060 (UDP or TCP) or 5061 (TLS).

Yes, with an IP allow list, allowed API keys, or both. Add the public addresses your equipment sends SIP from, or require each call to authenticate with an API key via SIP Digest. Both take effect on the next call.

The trunk's SIP domain stops accepting new calls immediately. Calls already in progress continue, and the call records made through the trunk stay in your call log.

Your PBX sends the call, Bird answers with a 407 challenge, and your PBX re-sends the call with a Proxy-Authorization header computed from username bird and your API key's secret as the password. Your PBX sends a hash of the password, never the password itself.

SHA-256 and MD5. Bird offers SHA-256 first and MD5 second by default, and your PBX picks the first one it supports. If your equipment only handles MD5 and mishandles a challenge that leads with SHA-256, set the trunk to MD5 only.

Yes. When a trunk has both, the source address is checked before Bird challenges for a password, so a call from an unlisted address is refused regardless of the credentials it carries.

Add the new key to the trunk first, move your equipment over, then revoke the old one. Revoking or deleting a key removes its ability to authenticate immediately on every trunk that allowed it.

A caller ID is a phone number your workspace is allowed to present as the caller on outbound calls. Bird checks the calling number your equipment puts in the SIP From header against this list on every call, using the eligible identities for your workspace, including qualifying numbers held in Bird and verified external numbers.

Add the number in international E.164 format and follow the verification flow. Answer the verification call and enter the code in the dashboard. Use the status shown on the number to confirm it is ready before calling.

If the attempts ran out, use Get a new code on the number's row for a fresh verification call. If the number is still waiting for its code, remove the caller ID and add the number again.

Arrange to answer the verification call through the system that receives calls for that number. If the number cannot receive the verification, contact support about the available ownership checks.

From that moment, any call presenting that number is refused with caller_id_not_verified. Calls already in progress continue, and the call records that used the number stay as they are.

Toll fraud works by dialing expensive countries you never meant to call. The countries you enable are the ones you can run up a bill in, so leaving everything else off limits your exposure if someone breaks into your phone system.

Find the country on the Destinations page using the search box (matches by name or two-letter code) and switch its toggle on. The change applies from that moment.

Calls to high-risk countries are expensive, and whoever operates the number you dial earns a share of the cost. These are the countries an attacker targets if they break into a phone system. Leave them off unless you have business there, and check the rate before you enable one.

Contact support to have it opened for your account. Available means you can enable it; Not supported means Bird cannot currently place calls to that country from your account.

An available country can still have specific destinations inside it that routing does not reach yet. Share the call ID with support to investigate destination coverage and the refusal.

Two headers: To (the number being called) and From (the number you present as the caller, which must be an eligible caller identity). Both must be full international numbers in E.164 format, a leading + followed by the country code and national number. No custom headers are needed.

STIR/SHAKEN lets providers sign calling identity information. The attestation reflects what the provider can verify about the caller and number. It does not guarantee the destination carrier will connect the call or remove a spam label.

Open the call in the call log. Your phone system sees a plain SIP 503, but the specific reason lands on the call record where only you can read it. A panel above the details names the cause and links to the setting that fixes it.

Only once the cause is gone. A call refused for calls_per_second_exceeded gets the same answer until you slow the dial rate. Read the rejection reason before retrying.

No. It means the call connected. Join a confirmed business event, such as a booking or resolved case, to evaluate task success.

An active call has not completed rating. Read the final record after the call and rating lifecycle finish; a missing value is not a zero price.

Verify the signature, deduplicate using the webhook identity and tolerate out-of-order arrival. Reconcile the final result against the call record.

Review direction and the report’s filters explicitly. The mix of traffic and the applicable rated components may differ.

Join rated calls and their sessions to confirmed outcomes in your application. Include every relevant call leg and use a consistent period and definition of completion.

Five: Answered (the called number picked up), No answer (it rang out), Failed (the call did not complete, either Bird refused it or a carrier failed it), Rejected (the carrier declined the call outright), and Unknown (the outcome could not be determined).

Both appear as Failed. The rejection reason is what separates them: only a Bird refusal carries one. A failed call with a rejection reason points at a setting on your side or Bird's; one without a rejection reason usually points at the number you dialed.

Yes. The Live tab on the Calls page lists the calls on your trunks right now, with a count. A live call shows as either Ringing (waiting for the other side to pick up) or In progress (connected). The tab refreshes every few seconds.

Total duration runs from the moment Bird received the call to hangup. Billable time runs from answer to hangup. The gap is ring time nobody picked up, so a large gap is worth a look at what you are dialing. Unanswered calls have no answered call time. Check the applicable rate card for other call components and charges.

Three ways: download CSV from the Calls page (exports every record matching your current filters, not only the visible page), read them over the API with an API key scoped to voice:read, or use the Bird CLI with bird voice list.

Three: voice_call.initiated (Bird accepted the call and started routing), voice_call.answered (the called number picked up), and voice_call.ended (the call is over, with the outcome). An unanswered call never emits the answered event.

Use the ended event for its reported status and durations, then reconcile the final outcome with the call record. Event publication can fail, and a request refused before admission can have no call record or events. A missing update does not establish that a call is still active.

Yes. Deliveries are not ordered, so answered can reach you after ended. Sort by the timestamp field, and let a later-arriving event with an earlier timestamp lose.

Deduplicate on the webhook-id HTTP header. A delivery or signaling retry can repeat an event. The same call and lifecycle stage retain their webhook identity. Reconcile a missing update against the call record.

They live on the call record, not in the event. A failed status on voice_call.ended does not tell you whether Bird or a carrier caused it. Open the call in the call log for the rejection reason, and the cost appears there once the call has been rated.

A call Bird cannot admit is turned away at the SIP layer, before a record exists. Check four things: the trunk has an IP range or API key allowing your equipment, the call arrived from an address on the trunk's IP allow list (behind NAT, that is the router's public address), the Digest credentials are correct (username bird, the right API key secret, an algorithm the trunk offers), and the SIP domain matches the trunk's domain exactly.

Open the call in the call log. The panel above the details names the cause and links to the setting that fixes it. The seven reasons you can fix are source_not_allowed, caller_id_not_verified, destination_not_enabled, insufficient_balance, daily_spend_exceeded, concurrent_calls_exceeded, and calls_per_second_exceeded.

Some equipment mishandles a challenge that leads with SHA-256. Set the trunk's Digest algorithm to MD5 only, and your PBX will get a challenge it understands.

Check the media addresses, firewall rules and supported codecs. A common cause is NAT sending media toward an unreachable private address. Configure the client’s NAT or STUN handling and verify audio in both directions.

It can be. Bird supports TLS on port 5061 for SIP signaling, so the call setup is encrypted in transit. UDP and TCP on port 5060 are unencrypted. Choose the transport your security requirements call for.

Every event is HMAC-signed. Verify the signature with your endpoint's secret before you act on the payload, and rotate that secret from the dashboard whenever you need to.

Voice records follow the configured service region. Review the applicable data-processing and retention terms for your deployment; API endpoint routing alone does not describe every processing or storage location.

Use voice access at write level for an allowed key to authenticate calls, and voice read access to read call records. The separate voice_management scope covers configuration. The trunk also applies its configured key and source rules.

The specific reason goes onto the call record where only you can read it. Returning a generic 503 to the SIP layer keeps someone probing your trunk from learning which trunks, numbers, and destinations exist.

Certifications and security documentation live in the Trust Center at trust.bird.com. The data processing agreement, privacy statement, and acceptable use policy are published at bird.com/legal. For a vendor questionnaire, your Bird account team handles it.

A call sequence controls prompts, keypad choices and recording inside a phone call. An automation flow coordinates a wider business process. Keep the two workflows and their publication decisions separate.

Yes. Combine a spoken or recorded prompt with keypad input and branches for recognized digits, timeout and unmatched input.

No. Keypad collection reads DTMF digits. A natural-language voice agent needs a conversational runtime with its own speech and turn-taking behavior.

Calls using a published version keep that version’s immutable content. New calls use the newly published version. Draft test calls read the latest saved sequence at each transition and can observe draft edits.

No. It identifies accepted work. Follow ringing, answer and the final call outcome separately, and keep uncertain retries tied to the same operation.

No. A phone menu follows predefined choices such as keypad digits. An AI voice conversation interprets language and generates responses through a conversational runtime. Both still need a complete call-routing and outcome-handling design.

A SIP-capable runtime can connect through Bird when its signaling, authentication and media behavior are compatible. Validate the integration with a real call in both directions before moving traffic.

No. MCP lets an authorized software agent operate account tools. A live voice agent additionally needs audio, speech processing, turn-taking and a conversation application.

Connect those actions through your application’s authorized tools. Treat their returned status as the source of truth, and require the appropriate customer confirmation before making a change.

Measure call connection and completion separately from task success. Track the confirmed business result, escalation rate and observed failure reasons; a transcript or an answered call alone cannot prove conversion.

No. Configure the number’s inbound route. A rejecting route is an explicit outcome, not an automatic connection to a team or agent.

Yes, through the call application or SIP runtime you configure to receive it. The number, inbound trunk, gateway and conversation runtime each need to be ready.

Inspect both correlated call legs. Check the verified target, enabled destination, available routing and applicable account limits.

No. The route records the selected handling behavior. The call status and related events tell you whether that behavior succeeded.

Forwarding creates an outbound leg as well as the arriving leg. Consult the published rates and the call’s final rated components.

No. The dashboard phone uses the browser and a configured Bird SIP trunk.

The dashboard phone places outbound calls. Route incoming calls to the phone system or application designed to answer them.

Yes. Use the keypad for DTMF choices such as an IVR menu or extension.

They are carried and rated as Voice calls. Review published rates for the destination and inspect the final call record.

No. Bird Voice does not carry emergency calls.

No. Recording is an explicit part of the configured workflow. Decide which supported source to capture and when capture starts and stops.

Audio capture and transcription are separate operations. Request transcription where appropriate and inspect its processing result.

Yes. Speech processing can finish after the telephone call. Keep call outcome and artifact processing state distinct.

No. A session can correlate several call legs, and each may have its own artifacts. Inspect the source call and capture identity.

Use it as conversation evidence. Confirm a booking, order change or payment in the business system that executed that action.